Alzheimer's disease is a result of beta-amyloid deposition and neurofibrillary tangles, which cause loss of neurons and their synapses, and, ultimately, atrophy of the brain. Studies have shown a relationship between oestrogen and Alzheimer's disease.Reference Lambert, Coyle and Lendon1 However, there is insufficient evidence to confidently prompt the use of oestrogen to reduce the risk of Alzheimer's disease.
Oestrogen, a reproductive and sex hormone, is produced by ovaries in women, as well as the adrenal glands and fat cells in small amounts. It plays a role in the formation of secondary sex characteristics during puberty, as well as for ovulation and thickening of the uterine wall in the menstrual cycle. Oestrogen also regulates cholesterol and glucose levels, synaptic plasticity in the brain Reference Rossetti, Cambiasso, Holschbach and Cabrera2 and other non-reproductive functions. In the menopause, ovaries produce less oestrogen, hence the use of hormone replacement therapy (HRT) to resupply oestrogen levels in post-menopausal women.
There is some evidence suggesting that oestrogen replacement therapy enhances spatial working memoryReference Lacreuse, Wilson and Herndon3 and cognitive functionReference Rapp, Morrison and Roberts4 in aged ovariectomised rhesus monkeys. A study using in vitro binding concluded that regulation of apolipoprotein E (APOE) induced by oestrogen increases the risk of Alzheimer's disease in women with an ε4 allele.Reference Lambert, Coyle and Lendon1 Another study utilising cultured basal forebrain neurons demonstrated that select oestrogens play a neuroprotective role.Reference Zhao and Brinton5 However, a study using PDAPP mice showed that oestrogen does not affect the deposition of beta-amyloid in the brain, thus having no effect on amyloid deposition in Alzheimer's disease.Reference Green, Bales, Paul and Bu6
Similarly, a randomised controlled trial (RCT) conducted in the USA highlighted the use of oestrogen replacement therapy in reducing beta-amyloid deposition.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 Additionally, findings from a cohort study suggest the duration of HRT use could be associated with the likelihood of Alzheimer's disease incidence in women.Reference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8 A case–control study revealed that HRT may prevent and delay the onset of Alzheimer's disease; moreover, findings showed that HRT might influence Alzheimer's disease risk when used during a critical window in early post-menopause.Reference Henderson, Benke, Green, Cupples and Farrer9
To date, it is unclear to what degree the association between oestrogen-only therapy has on the development of dementia, specifically Alzheimer's disease. Consequently, this systematic review is conducted to evaluate whether oestrogen replacement therapy has any significance in reducing the risk of Alzheimer's disease. In this paper, reducing risk is defined as including both prevention and delaying the onset of Alzheimer's disease.
Objective
This review aims to assess the effects of oestrogen replacement therapy on reducing the risk of Alzheimer's disease, and provide guidance on the future management of Alzheimer's disease with oestrogen replacement therapy. The hypothesis is that oestrogen replacement therapy is likely to reduce the risk of Alzheimer's disease in post-menopausal women.
Method
Details of the protocol for this systematic review were registered on PROSPERO (identifier CRD42022355335) and can be accessed at www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42022355335. The following are not publicly available: template data collection forms, data extracted from included studies, data used for all analyses and analytic code.
Criteria for considering studies for this review
All inclusion criteria were decided with the PICO framework, which focuses on population, intervention, comparison and outcomes to determine the different clinical components of this review.
Types of studies
A mix of double-blinded placebo RCTs, cohort studies and case–control studies were reviewed. All double-blinded placebo RCTs in which oestrogen replacement therapy or HRT was administered to women for at least 2 weeks, and cognitive function was measured, were eligible for inclusion. All cohort studies in which oestrogen replacement therapy or HRT was recorded to have been given to women, and cognitive function was measured or diagnosis of Alzheimer's disease was made, were eligible for inclusion. All case–control studies in which oestrogen replacement therapy or HRT was recorded to be administered to women, and cognitive function was measured or diagnosis of Alzheimer's disease was made, were eligible for inclusion.
Types of participants
Participants in RCTs were required to be women of any age, be peri- or post-menopausal, have an intact uterus or have had a hysterectomy, and not have been diagnosed with Alzheimer's disease. Participants in cohort studies were required to be peri- or post-menopausal women (naturally or surgically induced), with an intact uterus or having undergone a hysterectomy, of any age, with suspected or diagnosed Alzheimer's disease. Participants in case–control studies were peri- or post-menopausal women with an intact uterus or having undergone a hysterectomy, of any age, with suspected or diagnosed Alzheimer's disease.
Types of interventions
Interventions containing oestrogens alone (oestrogen replacement therapy) or combined with a progestogen (HRT) at any dose, any dosing schedule and any mode of administration (oral, subdermal, transdermal or intravenous), were eligible for inclusion.
Types of outcome measures
The primary outcome of interest is cognitive function, as measured either by validated neuropsychological tests (some of the tests have multiple functions and may span other categories to the one in which they have been placed) or by strict protocols.
Tests used for overall global cognitive function were the Modified Mini-Mental State Examination (3MSE), Mini-Mental State Examination (MMSE), Cambridge Mental Disorders of the Elderly Examination (CAMDEX), Clinical Dementia Rating (CDR) Scale, Bi-factor Model for the Cognitive Baseline Data, Global Deterioration Scale and Alzheimer's Disease Assessment Scale (ADAS-Cog).
Cognitive impairment (defined and assessed by strict protocols) was defined according to the DSM-III, DSM-III-R and DSM-IV; the National Institute of Neurological and Communicative Disease and Stroke/Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) criteria and the ICD-10.
Specific tests used for individual cognitive domains were categorised as verbal memory and language tests (measured with the California Verbal Learning Test (CVLT) and Wechsler Memory Scale (WMS)), visuospatial tests (measured with the Benton Visual Retention Test and the WMS) and executive function tests (measured with the WMS). WMS includes portions of both the WMS and WMS Revised: logical memory subtest, visual reproductions subtest, forward and backward visual memory spans, and forward and backward digit spans.
Magnetic resonance image (MRI) scans were obtained with a 1.5 T Siemens Magnetom Vision scanner with a standard radio frequency head coil. T1-weighted images with 1 mm isotropic resolution was acquired.
Outcomes for cohort and case–control studies included questionnaires, interviews and death certificates to collate additional information, including total lifetime oestrogen exposure (age of menarche, age of menopause, natural or surgical menopause, parity, lifetime breastfeeding history, HRT history, type of oestrogen preparation, dosage and oral contraceptive use), vascular pathology (history of myocardial infarction, hypertension and hypercholesterolaemia) and education level.
Search methods for identification of studies
This systematic review targeted articles published within the past 30 years (1992–2022), as we believe this time frame would adequately allow us to find the minimum number of RCTs (four) needed to increase the reliability of our study and provide the most relevant data for our evaluation.
Through the use of healthcare databases available to us, our selections were narrowed down to RCTs, cohort studies and case–control studies, as the three offered higher levels of evidence for our systematic review. Among the databases we have listed below, Ovid was used to access Medline and PsycINFO, and EBSCO was used to access EMBASE and CINAHL. Each source was last searched in July 2022.
Studies evaluated in this systematic review had an inclusion criteria of English language publication; the objective of reducing risk, delaying onset or prevention; and data involving peri-menopausal and post-menopausal women. Exclusion criteria include diagnosis of dementia other than Alzheimer's disease, progesterone-only therapy and pre-menopausal women.
The following are the list of healthcare databases and the general search terms used (full list in Supplementary Appendix 1 available at https://doi.org/10.1192/bjo.2023.579): PubMed, Science Direct, Scopus, BMJ, EMBASE (EBSCO), Web of Science, CINAHL (EBSCO), PsycINFO (Ovid) and The Cochrane Central Register of Controlled Trials; search terms ‘oestrogen therapy’ OR ‘estrogen therapy’ OR ‘hormone replacement therapy’ and ‘Alzheimer's disease’ OR ‘Alzheimer's dementia’.
Selection of studies and data collection
Studies were selected based on their abstracts. All reviewers read the abstracts (G.R.M.W., H.R., Q.Y.L. and E.J.A.L.) to screen for eligible studies in the review (RCTs, cohort studies and case–control studies). Further reading was done on each study, and any disparities in opinion were discussed and finalised among all reviewers. The selection of studies was made based on the reviewers’ definition of ‘reducing risk’, which includes ‘prevention’ and ‘delaying onset’. All reviewers performed data collection from included papers. Data were extracted on 2 September 2022.
Description of studies
Fifty-nine studies were initially identified and discussed among the reviewers (Fig. 1). After exclusion of studies that were not RCTs, cohort studies or case–control studies, a total of 41 studies remained. A final discussion was held after a thorough reading of the remaining studies, and 24 studies were selected after matching the criteria set by the reviewers.
Fig. 1 Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) flow diagram of search process and identification of studies.
Of the 24 studies, there were four RCTs, nine cohort studies and 11 case–control studies. Exposure to oestrogen replacement therapy ranged from 16 weeks to 8 years. Twenty studies did not have data on treatment intervention duration because they utilised previous use of oestrogen as a measure for oestrogen replacement therapy.
Participants
Screening and selection
As this systematic review includes RCTs, cohort studies and case–control studies, the total number of participants differs by the type of study. The total number of participants in RCTs is on a smaller scale; for instance, 118Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 or 42Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 participants, whereas another had 4532 participants for their clinical trialReference Shumaker, Legault and Rapp11 (Table 1). In comparison, the largest number of participants was 4 696 633 in a nationwide study,Reference Yoo, Shin, Han, Kim, Won and Lee12 and the remaining cohort studies ranged from 688Reference Paganini-Hill and Henderson13 to 5092Reference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8 (Table 2). The sample size used in case–control studies also varied, ranging from 280Reference Seshadri, Zornberg, Derby, Myers, Jick and Drachman14 to 230 580Reference Imtiaz, Taipale, Tanskanen, Tiihonen, Kivipelto and Heikkinen15 (Table 3). This trend is because of the aim, duration and nature of each study.
Table 1 Summary of randomised controlled trials
ITT, intention to treat; CEE, conjugated equine oestrogen; APOE, apolipoprotein E.
a. Data to calculate relative risk was not available.
b. Calculated based on the treatment group utilising oestrogen therapy only.
Table 2 Summary of cohort studies
HRT, hormone replacement therapy; CEE, conjugated equine oestrogen.
a. Unable to contact authors for further clarification.
b. Note that this study is a different study from the Imtiaz et alReference Imtiaz, Taipale, Tanskanen, Tiihonen, Kivipelto and Heikkinen15 case–control study.
c. Data to calculate odds ratio was not available.
Table 3 Summary of case studies
HRT, hormone replacement therapy; CEE, conjugated equine oestrogen.
a. Note that this study is a different study from the Imtiaz et alReference Imtiaz, Tuppurainen, Rikkonen, Kivipelto, Soininen and Kröger25 cohort study.
b. Unable to contact authors for further clarification.
All except oneReference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8 of the studies included in this review only involved women, specifically peri- or post-menopausal women, and each study stated its criteria. For example, participants with existing neurological or psychiatric illnesses,Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7,Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16–Reference Slooter, Bronzova, Witteman, Van Broeckhoven, Hofman and van Duijn19 or other types of dementia,Reference Paganini-Hill and Henderson13 were excluded. Because of the role and interaction of oestrogen with other reproductive and cardiovascular factors, participants who had cancer within a specified time range,Reference Kim, Soto, Branigan, Rodgers and Brinton20 or had either an oophorectomy or hysterectomy,Reference Yoo, Shin, Han, Kim, Won and Lee12, Reference Henderson, Paganini-Hill, Emanuel, Dunn and Buckwalter21 or stroke,Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16 were excluded. Since RCTs required administration of oestrogen replacement therapy, participants who used oestrogen or progesterone before the time of initiation, or had a previous diagnosis of dementia, were excluded from analysis;Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7,Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 two studies even excluded participants with high risk of developing dementia at baseline.Reference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22
Confounding factors (age, mood and others)
Most studies included participants from a wide age range with a minimum age of 50–60 years (except for the study by Kim et alReference Kim, Soto, Branigan, Rodgers and Brinton20), in an attempt to standardise age effects. A history of mental health illness is said to influence study results. Therefore, to standardise results, some studies excluded patients with psychiatric disorders or those receiving antipsychotic or antidepressant medications.Reference Lord, Buss, Lupien and Pruessner17,Reference Brenner, Kukull, Stergachis, van Belle, Bowen and McCormick18 Education and income level was mentioned in multiple studies, as it affects access to healthcare. One studyReference Lord, Buss, Lupien and Pruessner17 specifically mentioned that all participants were controlled so that there was no obvious discrepancy in education level, depressive episodes, body mass index, exercise level, alcohol intake, smoking habits and income level, as these were identified to possibly influence results.
Interventions
A variety of methods were used in administration of oestrogen replacement therapy, dosage and types of oestrogen replacement therapy, as well as whether progesterone was added.
All RCTs used Premarin, a conjugated equine oestrogen (CEE), and a placebo. Two of the RCTs used a dosage of 0.625 mg per day of Premarin.Reference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 One studyReference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 used 1.25 mg per day of Premarin. One RCTReference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 used either 0.45 mg per day of Premarin or 50 μg per day of Climara (oestradiol). All four RCTs administered oestrogen via oral tablet, including one that utilised both transdermal patches and oral tablets. Of the four RCTs, all co-administered progesterone together with oestrogen replacement therapy. One studyReference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 used 200 mg per day of Prometrium (micronized progesterone) orally for the first 12 days of each month in their trial. In one study,Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 progesterone was given only if participants had an intact uterus.
Among the RCTs, one studyReference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 did not list any adherence checks and three studiesReference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10,Reference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 stated that they checked for participant adherence without stating the methods used.
Three cohort studiesReference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8,Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16,Reference Kawas, Resnick, Morrison, Brookmeyer, Corrada and Zonderman23 used medical histories and three studiesReference Yoo, Shin, Han, Kim, Won and Lee12,Reference Paganini-Hill and Henderson13,Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 used surveys to categorise participants with lifetime use of oestrogen therapy. Among these studies, fourReference Yoo, Shin, Han, Kim, Won and Lee12,Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16,Reference Kawas, Resnick, Morrison, Brookmeyer, Corrada and Zonderman23,Reference Imtiaz, Tuppurainen, Rikkonen, Kivipelto, Soininen and Kröger25 categorised participants based on the duration of oestrogen use.
Six case–control studies accessed participant information via medical records, three studiesReference Henderson, Benke, Green, Cupples and Farrer9,Reference Brenner, Kukull, Stergachis, van Belle, Bowen and McCormick18,Reference Henderson, Paganini-Hill, Emanuel, Dunn and Buckwalter21 accessed information through medical interviews and only one studyReference Paganini-Hill and Henderson26 accessed information through a survey. One studyReference Slooter, Bronzova, Witteman, Van Broeckhoven, Hofman and van Duijn19 collected information from the next of kin through structured interviews. Two studiesReference Roberts, Cha, Knopman, Petersen and Rocca27Reference Pourhadi, Mørch, Holm, Torp-Pedersen and Meaidi28 categorised their participants based on the initial age of initiation of oestradiol tablets. One studyReference Imtiaz, Taipale, Tanskanen, Tiihonen, Kivipelto and Heikkinen15 categorised their participants based on years of oestrogen use, progestogen use and combined HRT use. Five studiesReference Seshadri, Zornberg, Derby, Myers, Jick and Drachman14,Reference Kim, Soto, Branigan, Rodgers and Brinton20,Reference Roberts, Cha, Knopman, Petersen and Rocca27,Reference Waring, Rocca, Petersen, O'Brien, Tangalos and Kokmen29,Reference Kim, Soto, Branigan, Rodgers and Brinton30 categorised their participants based on the previous duration of use of oestrogen. One study compared a group of patients with early-onset Alzheimer's disease and a group of patients with none of the symptoms of dementia.Reference Slooter, Bronzova, Witteman, Van Broeckhoven, Hofman and van Duijn19
Quality assessment
The overall quality of evidence for all included RCTs were assessed with the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) assessment (Supplementary Table 4). Quality of cohort studies and case–control studies were assessed with the Newcastle–Ottawa Scale, of which the studies included were classified as ‘good’ (n = 13) or ‘fair’ (n = 7) (Supplementary Table 5). Quality of evidence was assessed by all four reviewers.
Results
RCTs
Outcomes
Cognitive assessments
The RCTs assessed the cognitive functions of participants by using specific tests for individual cognitive domains. However, each trial used different cognitive tests with little overlap, thus making the pooling of data for analysis difficult. It was noted that distinct aspects of the same tests or different versions of the same test were used in different studies. Analysis was split for tests of verbal and visual memory, and for those that tested other aspects of cognitive function.
Two studiesReference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 used the same global measure to assess cognitive function, the 3MSE. In addition, both studies analysed the incidence of mild cognitive impairment among participants and compared the treatment group with a placebo group.
Statistics
Some studiesReference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10,Reference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 analysed baseline and post-treatment effects separately. Because of the increased number of comparisons, adjustment is required to reduce the incidence of chance findings.
Risk of bias in included studies
The RCTs included were all double-blinded and placebo-controlled.
Randomisation
Three of the studies provided details on their randomisation methods,Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10,Reference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 whereas one study did not.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7
Allocation concealment
One study was clear on their method of allocation concealment;Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 however, the remaining three studies had inadequate description of their method for allocation concealment.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7,Reference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 There was also presumably suspected allocation to active treatment in one study.Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10
Attrition
All studies lost a variable number of participants throughout their trials. One studyReference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 excluded 42.4% of participants from analysis. Most of the participants excluded from the analysis were from the CEE treatment group, with only 43.6% included.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 This was largely because of non-participation in their MRI and positron emission tomography scans. In two studies,Reference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 the attrition was 0%, as the authors included all participants in the data analysis despite losing some participants because of incomplete data and other reasons.
Intention-to-treat analyses
Two studiesReference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 described the use of intention-to-treat analyses performed. However, the remaining two studies had no information on any intention-to-treat analysis performed.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7,Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10
Power calculation for sample size
Two studiesReference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 undertook power calculations for their sample size based on the protocols of the Women's Health Initiative Memory Study (WHIMS),Reference Shumaker, Reboussin, Espeland, Rapp, McBee and Dailey31 which estimates at least 80% power to detect a difference of 40%. The other studies did not disclose any information on performing power calculations for their trials.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7,Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10
Effects of interventions
Test of overall global cognitive function
One study used the 3MSE to screen for global cognitive impairment and track changes in global cognitive function.Reference Shumaker, Legault and Rapp11 The cut-off point for the study was a 3MSE score of ≤80 points for participants with ≤8 years of education and ≤88 points for participants with ≥9 years of education. The main effects for baseline 3MSE scores alone were statistically significant (P < 0.001). Results showed that the risk of developing probable dementia was 3.78 times (95% CI 1.91–7.50) greater for women with baseline 3MSE scores ranging from cut-off to 94 points; and 24.48 times (95% CI 13.19–46.75) greater for women with baseline 3MSE scores at or below cut-off compared with women with 3MSE scores ranging from 95 to 100.
In another study, the 3MSE was used similarly, and the cut-off point remained the same.Reference Imtiaz, Taipale, Tanskanen, Tiihonen, Kivipelto and Heikkinen15 Results showed that combining both the oestrogen-only and the oestrogen plus progestin trials showed an overall hazard ratio of 1.76 (95% CI 1.19–2.60; P = 0.005). However, after excluding participants with baseline 3MSE scores at or below the cut-off point, the hazard ratio was 1.77 (95% CI 0.74–4.23; P = 0.20) in the oestrogen-only trial.
One studyReference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 used the ADAS-Cog scale and one studyReference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 used a bi-factor model adapted from Dowling et alReference Dowling, Gleason, Manson, Hodis, Miller and Brinton32 to measure outcomes for change in global cognitive function. One study showed no significant correlation between global cognitive scores and beta-amyloid deposition in the brain, measured by Pittsburgh compound-B standard unit value ratio.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 The remaining trialReference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 found no significant differences in participant groups with global cognitive function measures.
Mild cognitive impairment
One study found that the risk of being diagnosed with mild cognitive impairment was not statistically significant between the two groups (hazard ratio 1.07, 95% CI 0.74–1.55; P = 0.72).Reference Shumaker, Legault and Rapp11 However, the risk of being diagnosed with mild cognitive impairment or probable dementia increased by 37% for women in the exposure group who were receiving oestrogen plus progestin (hazard ratio 1.37, 95% CI 0.99–1.89; P = 0.06).Reference Shumaker, Legault and Rapp11
Another study showed that the risk of being diagnosed with mild cognitive impairment was 34% higher in the CEE group than in the placebo group (hazard ratio 1.34, 95% CI 0.95–1.89).Reference Shumaker, Legault and Kuller22 Likewise, the risk was similar in the combined oestrogen and progestin trial (hazard ratio 1.25, 95% CI 0.97–1.60).Reference Shumaker, Legault and Kuller22 Neither of the hazard ratios were statistically significant.
Specific tests for individual cognitive domains
Verbal memory and language tests
No association was noted in one studyReference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 between oral CEE therapy and beta-amyloid deposition compared with placebo. However, after adjusting for age, education, time from menopause to randomisation and APOE ɛ4 status, the CVLT total score was lower in the oral CEE group compared with the placebo group (P = 0.03).
One study,Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 showed overall insignificant results, with only delayed paragraph recall showing a positive result at 16 weeks for the CEE group.
Visuospatial tests
Two studies did not include details on visuospatial tests. One study tested visual memory with the Benton Visual Retention Test, but the results were unavailable.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 Another studyReference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 showed a significant difference on the visual memory span forward test at week 16.
Speed tests, attention and manual dexterity, and semantic memory tests
Outcome measures for speed tests, attention tests and semantic memory tests showed no statistically significant differences in all of the studies. Some studiesReference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10,Reference Shumaker, Legault and Rapp11,Reference Shumaker, Legault and Kuller22 did not include specific details on the P-values of each test carried out.
Executive function
All studies performed outcome measures for executive function. In one study,Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 a statistically significant difference was found in the Trail Making Test Part B, with a P-value of <0.05. In the same study, the backward digit span test found a near statistically significant difference, with a P-value of between 0.05 and 0.1. However, the authors did not disclose the exact value of the test. All other studies found no statistically significant difference in the other tests that measured executive function.
Cohort studies
Outcomes
Cognitive assessments
Most of the cohort studies utilised widely used cognitive assessment tools that mainly assessed global cognitive functions. For instance, the MMSE, CDR and Global Deterioration Scale were used in one study.Reference Yoo, Shin, Han, Kim, Won and Lee12 Other than the MMSE, DSM-III-R and NINCDS-ARDA, one studyReference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 used an extensive list of cognitive assessments, such as the CAMDEX and ICD-10 criteria for other dementia diseases. One studyReference Lord, Buss, Lupien and Pruessner17 used MRI scans to assess integrity of the amygdala and hippocampus to investigate the neuroprotective effects of HRT.
One studyReference Paganini-Hill and Henderson13 that used a case–control methodology within a prospective cohort study conducted their study using death certificates to confirm the diagnosis of Alzheimer's disease, therefore no cognitive assessments were performed in the course of their study.
Statistics
Although most studies investigated various reproductive factors, some chose Cox proportional hazards regression analysis to accommodate multiple covariates,Reference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8,Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16,Reference Kawas, Resnick, Morrison, Brookmeyer, Corrada and Zonderman23,Reference Imtiaz, Tuppurainen, Rikkonen, Kivipelto, Soininen and Kröger25 whereas others used multivariate regression analysis.Reference Paganini-Hill and Henderson13,Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 Studies that performed age-specific comparisons utilised the combination of analysis of variance (ANOVA) for continuous variables and chi-squared for categorical variables;Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16,Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 one study used the combination of ANOVA for continuous and Student's t-test for categorical variables.Reference Mortel and Meyer33 Other characteristics of those who did and those who did not use HRT were analysed with chi-squared tests for categorical variables and two-sample t-tests for continuous measures.Reference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8,Reference Imtiaz, Tuppurainen, Rikkonen, Kivipelto, Soininen and Kröger25
Risk of bias in included studies
Attrition
One study had an attrition rate of 27.3%,Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 mainly because of an inability to obtain a history of oestrogen therapy, drop-out rate and other unspecified reasons. The trial with the highest attrition rate of 42.4%Reference Imtiaz, Tuppurainen, Rikkonen, Kivipelto, Soininen and Kröger25 reported lack of response, changes in address and death as the main causes of attrition. Some studiesReference Yoo, Shin, Han, Kim, Won and Lee12,Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16,Reference Kawas, Resnick, Morrison, Brookmeyer, Corrada and Zonderman23 had comparatively lower rates of attrition of <10%. One studyReference Mortel and Meyer33 did not disclose the attrition rate in their study.
Because of the scale and duration of some studies, follow-ups could not be performed because of reasons such as participants passing away, missing data over the course of the study, refusal to participate in the study and inability to locate participants.Reference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8,Reference Yoo, Shin, Han, Kim, Won and Lee12,Reference Paganini-Hill and Henderson13,Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24,Reference Imtiaz, Tuppurainen, Rikkonen, Kivipelto, Soininen and Kröger25 This shows the possibility of bias, as the loss of participants to follow-up potentially affects the overall results. Furthermore, as most of the studies involved older women, attrition highlights the particular bias resulting from death and decline in the health of participants, which leads to potential bias of data toward those who are healthy enough to participate until the end of the study.
Test of overall global cognitive function
Two studiesReference Yoo, Shin, Han, Kim, Won and Lee12,Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 measured cognitive function with the MMSE. They had a cut-off score of 26 and 23/24 out of 30 to indicate dementia for each of the studies, respectively. Significantly more women who had never used HRT were diagnosed with Alzheimer's disease in these two studies. One studyReference Yoo, Shin, Han, Kim, Won and Lee12 used the CDR to diagnose new-onset dementia after a median of 5.73 years from baseline. In contrast, another study showed that women using HRT had lower risks and delayed onset of dementia.Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 Another studyReference Tang, Jacobs, Stern, Marder, Schofield and Gurland16 showed that the age at onset for Alzheimer's disease is significantly later in women who have used oestrogen replacement therapy compared with those who have not (log-rank test P < 0.01). It also showed that adjustments made for possible confounding factors, such as ethnic group, education and participation group, did not significantly affect the relative risk (relative risk 0.5, 95% CI 0.25–0.9; P = 0.02).
Five studiesReference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8,Reference Kawas, Resnick, Morrison, Brookmeyer, Corrada and Zonderman23–Reference Imtiaz, Tuppurainen, Rikkonen, Kivipelto, Soininen and Kröger25,Reference Mortel and Meyer33 compared the results from different cognitive assessment tests to specific criteria such as the NINCDS-ADRDA criteria for probable Alzheimer's disease and DSM-III-R criteria for dementia syndrome. They all yielded positive results favouring the use of hormone therapy in delaying the onset and prevention of dementia, with one studyReference Kawas, Resnick, Morrison, Brookmeyer, Corrada and Zonderman23 highlighting the benefits of long-term HRT use.
Case–control studies
Outcomes
Cognitive assessments
One studyReference Waring, Rocca, Petersen, O'Brien, Tangalos and Kokmen29 measured Alzheimer's disease according to DSM-III criteria; however, not all of their patients received routine neuropsychological testing. Two studiesReference Imtiaz, Taipale, Tanskanen, Tiihonen, Kivipelto and Heikkinen15,Reference Roberts, Cha, Knopman, Petersen and Rocca27 measured Alzheimer's disease according to DSM-IV criteria. Two studiesReference Seshadri, Zornberg, Derby, Myers, Jick and Drachman14Reference Imtiaz, Taipale, Tanskanen, Tiihonen, Kivipelto and Heikkinen15 used the NINCDS-ADRDA criteria to measure for probable Alzheimer's disease. Two studiesReference Brenner, Kukull, Stergachis, van Belle, Bowen and McCormick18,Reference Henderson, Paganini-Hill, Emanuel, Dunn and Buckwalter21 made use of the MMSE as part of their neuropsychological testing to measure dementia severity. One studyReference Pourhadi, Mørch, Holm, Torp-Pedersen and Meaidi28 used the criteria for Alzheimer's disease set by ICD-10. Another studyReference Slooter, Bronzova, Witteman, Van Broeckhoven, Hofman and van Duijn19 used NINCDS-ARDA criteria, the CDR and a short portable mental status questionnaire to assess any progressive decline of intellectual functions during the course of study.
Statistics
One studyReference Henderson, Paganini-Hill, Emanuel, Dunn and Buckwalter21 used a general linear modelling programme to control for potential confounding factors such as age and education, and used the extension chi-squared procedure to remove the controlled variables. Another trialReference Henderson, Benke, Green, Cupples and Farrer9 evaluated confounding factors by comparing crude odd ratios with Mantel–Haenszel odds ratios. There was use of multivariable models and subsequent evaluation in indirect analysis to examine confounding variables in another study.Reference Roberts, Cha, Knopman, Petersen and Rocca27 Logistic regression models were also used with regards to matching in some studies,Reference Seshadri, Zornberg, Derby, Myers, Jick and Drachman14,Reference Imtiaz, Taipale, Tanskanen, Tiihonen, Kivipelto and Heikkinen15 Reference Kim, Soto, Branigan, Rodgers and Brinton20Reference Waring, Rocca, Petersen, O'Brien, Tangalos and Kokmen29 whereas another studyReference Roberts, Cha, Knopman, Petersen and Rocca27 discounted it to reduce loss of statistical power caused by missing data. Finally, a study that included the APOE population conducted stratified analysis by using unconditional logistic regression to test for statistical interaction between oestrogen use and the APOE genotype.Reference Slooter, Bronzova, Witteman, Van Broeckhoven, Hofman and van Duijn19
Risk of bias in included studies
Attrition
Two studiesReference Kim, Soto, Branigan, Rodgers and Brinton20,Reference Waring, Rocca, Petersen, O'Brien, Tangalos and Kokmen29 had no information on the attrition in their studies. Most studies had <10% attrition, threeReference Seshadri, Zornberg, Derby, Myers, Jick and Drachman14,Reference Imtiaz, Taipale, Tanskanen, Tiihonen, Kivipelto and Heikkinen15,Reference Henderson, Paganini-Hill, Emanuel, Dunn and Buckwalter21 of which had an attrition rate of 0%. One studyReference Pourhadi, Mørch, Holm, Torp-Pedersen and Meaidi28 had an attrition rate of 78.1% because of censoring during follow-up, which occurred for a multitude of reasons including, but not limited to, emigration and missing information on socioeconomic confounders.
Test of overall global cognitive function
There was no significant overall benefit of oestrogen replacement therapy and HRT on Alzheimer's disease seen in the case–control studies, with only half of them showing positive results.Reference Henderson, Benke, Green, Cupples and Farrer9,Reference Kim, Soto, Branigan, Rodgers and Brinton20,Reference Henderson, Paganini-Hill, Emanuel, Dunn and Buckwalter21,Reference Paganini-Hill and Henderson26,Reference Waring, Rocca, Petersen, O'Brien, Tangalos and Kokmen29 The other trials all showed little to no association of oestrogen replacement therapy with Alzheimer's disease. One trialReference Waring, Rocca, Petersen, O'Brien, Tangalos and Kokmen29 showed improved cognition for patients that used HRT for a longer duration, at least 6 months (odds ratio 0.42, 95% CI 0.18–0.96; P = 0.04). Another trialReference Paganini-Hill and Henderson26 showed a decreased risk of Alzheimer's disease with both an increased dosage and duration of oestrogen replacement therapy (odds ratio 0.65, 95% CI 0.49–0.88; P = 0.01). One studyReference Slooter, Bronzova, Witteman, Van Broeckhoven, Hofman and van Duijn19 used the CDR to screen new participants for dementia.
Discussion
Out of all of the studies analysed in this systematic review, some studies show positive associations in oestrogen replacement therapy reducing risk of Alzheimer's disease (n = 15), and the remaining studies found no evidence of such an association (n = 9). One RCT associated transdermal 17β-oestradiol with reducing the progression of Alzheimer's disease pathology in the population of APOE ɛ4 carriers,Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 whereas three other studies showed no association between short-term oestrogen replacement therapy and Alzheimer's disease. Cohort and case–control studies had a similar number of positive associations. In conclusion, the evidence of oestrogen replacement therapy in reducing risk of Alzheimer's disease remains inconsistent; however, further details regarding oestrogen's role in Alzheimer's disease were discovered.
Some studies investigated oestrogen-related variables, such as age of menarche, date of most recent menstrual period and number of pregnancies, and found that risk for Alzheimer's disease increased with shorter oestrogen exposure.Reference Yoo, Shin, Han, Kim, Won and Lee12,Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 This principle seemed to be applied similarly with the time of HRT exposure. Longer use of oestrogen replacement therapy is associated with lower risk of developing Alzheimer's disease,Reference Henderson, Benke, Green, Cupples and Farrer9 whereas women who had never used oestrogen replacement therapy had 65% risk of the disease.Reference Paganini-Hill and Henderson13 One case study, with a mean treatment duration of 3.1 years, found that those who received oestrogen replacement therapy had a higher MMSE score of 16.6, compared with an MMSE score of 9.9 for those who did not receive oestrogen replacement therapy (P = 0.02).Reference Henderson, Paganini-Hill, Emanuel, Dunn and Buckwalter21 Further, despite the smaller effect, a cohort study found that women taking oestrogen replacement therapy had better MMSE scores (25.9 v. 24.2; P < 0.001).Reference Baldereschi, Di Carlo, Lepore, Bracco, Maggi and Grigoletto24 Duration of treatment and types of treatment are believed to heavily influence cognitive outcomes,Reference Ryan, Carriere, Scali, Dartigues, Tzourio and Poncet34 both of which differed in both studies. The difference may also be attributable to women in the two trials differing in many factors at baseline, including cognitive function.
Two studies that evaluated the effects of oestrogen replacement therapy on five domains of cognitive function resulted in significant findings on ‘learning and memory’, whereas no evidence of significant effect was found in any other domains. One study concluded with no correlations between oestrogen replacement therapy and cognitive function measured with the CVLT,Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 whereas another study found that verbal memory performances (measured via logical memory test) were better after 16 weeks of oestrogen replacement therapy.Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10 A possible explanation for this contrasting conclusion could be the use of different cognitive tests; for instance, a study investigated the differences between several neuropsychological assessments, including the CVLT and logical memory test, and discourages using the two interchangeably because there is a high chance of generating a discrepant score.Reference Thiruselvam, Vogt and Hoelzle35 Similarly, a meta-analysis comprising 24 papers revealed that most trials showed significant findings in ‘learning and memory’, suggesting that this domain is the most susceptible to the effects of oestrogen.Reference Lethaby, Hogervorst, Richards, Yesufu and Yaffe36 It is believed that this is likely because of an interaction between oestrogen and the cholinergic neurotransmitter systems on the function of hippocampus and prefrontal cortex, affecting the main stage of memory processing.Reference Maki and Dumas37
The APOE genotype is the most common genetic determinant of risk of Alzheimer's disease. Several studies included additional investigation of APOE ɛ4 carriers who have greater risk of Alzheimer's disease. One studyReference Tang, Jacobs, Stern, Marder, Schofield and Gurland16 suggested reduced risk of Alzheimer's disease in women heterozygous for ɛ4 with oestrogen replacement therapy use, whereas another studyReference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8 claimed that two ɛ4 alleles had a greater risk reduction, although the author was uncertain of this finding. This highlights the differences in mechanisms of cognitive decline and pathology in APOE ɛ4 carriers compared with non-carriers. According to Yaffe, oestrogen may play a role against carotid atherosclerosis by interacting with APOE ε4, as cardiovascular problems have been proven to result in Alzheimer's disease.Reference Yaffe38 This was similar to a hypothesis by Depypere et al, who additionally highlighted the importance of looking for the ‘window of opportunity’ to see a maximal effect of oestrogen replacement.Reference Depypere, Vierin, Weyers and Sieben39 Yaffe also proposed a possible modulation of acetylcholine in the brain, resulting in increased neuronal growth and synaptic reorganisation as a result of oestrogen's effect in regulating synaptic plasticity through stimulation of axonal sprouting and dendritic spine formation, which is evident in studies carried out on rodent tissues. This is also reiterated by Stone et al, who demonstrated that oestrogen causes changes in synaptic sprouting through an APOE-mediated mechanism in response to injury in mice.Reference Stone, Rozovsky, Morgan, Anderson and Finch40 Stone et al also hypothesised that the increased risk of Alzheimer's disease in female APOE ε4 carriers could be a result of a combination of effect between oestrogen-deficient state and ability for neuronal reorganisation and choline acetyltransferase activity through the ε4 genotype. On the other hand, Depypere et al also proposed that APOE plays a role in increasing potential neuroprotective actions of oestrogen. This is thought to be a result of lipoprotein receptor-related protein, which is an APOE receptor, where expression after stimulation by oestrogen causes an increased amount of APOE produced by astrocytes. The produced APOE then stimulates an increase in lipids, which is necessary for cell membrane growth, ultimately resulting in neuronal growth. A trial conducted by Kantarci et al, using transdermal oestrogen on APOE carriers, found that oestrogen also plays a role in reducing amyloid plaque formation.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 This was further solidified by a study done on primary cultures derived from rodent and human neocortex.Reference Xu, Gouras, Greenfield, Vincent, Naslund and Mazzarelli41
Future studies should explore the critical time window of HRT intervention, as some studies have suggested early menopause to be the optimal period for HRT to play an effective role in delaying dementia progression.Reference Henderson, Benke, Green, Cupples and Farrer9 It will be interesting to see whether this critical time period has any associations with different ethnicities or subgroups of patients who are APOE ɛ4 carriers or non-carriers.
Two studies investigated the concomitant use of non-steroidal anti-inflammatory drugs, one of which suggested a potential role in neuroprotection of Alzheimer's disease.Reference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8 In contrast, the other study had inconclusive results because of a lack of data.Reference Kawas, Resnick, Morrison, Brookmeyer, Corrada and Zonderman23 This variable has yet to justify a role in oestrogen replacement therapy and Alzheimer's disease. The proposed mechanism of how non-steroidal anti-inflammatory drugs decrease beta-amyloid 1-42 production is by allosteric modulation of gamma-secretase activity, the enzyme responsible for the formation of beta-amyloid.Reference Imbimbo42 The potential concomitant use of non-steroidal anti-inflammatory drugs not only provides new understanding of the pathogenesis of Alzheimer's disease, but has also created opportunities to explore other potential preventive pharmacologic agents for reducing Alzheimer's disease risk.
A strength of this review is that it includes papers published recently,Reference Kim, Soto, Branigan, Rodgers and Brinton20 which suggests that this topic remains an ongoing debate. Also, only RCTs, cohort studies and case–control studies are included in our analysis, to provide more credible and quality analysis. Finally, a selection of oestrogen products, multiple routes of administration (oral, transdermal, vaginal, injections) and different durations of oestrogen replacement therapy were analysed in the papers reviewed.
A limitation of the studies used is the duration of oestrogen treatment. Trials conducted between 16 weeks and 12 months found no associations, and suggested that the trial period was unlikely to have significant results.Reference Henderson, Paganini-Hill, Miller, Elble, Reyes and Shoupe10,Reference Shumaker, Legault and Rapp11 From this, clinical trial periods between 12 months and 4 years are highly recommended, as none of the papers reviewed included this time frame. Furthermore, since most RCTs use the same type of oestrogen replacement therapy, dose and route in their trials, the data found may not be applicable to other scenarios or the general population. A combination of different routes, dosage or type of oestrogen replacement therapy explored in a single RCT might be helpful to uncover new findings.
As surveys or interviews are used to gather information in cohort and case–control studies, data collected are based on participants’ memories, which could be inaccurateReference Slooter, Bronzova, Witteman, Van Broeckhoven, Hofman and van Duijn19 and may be incomplete or insufficient.Reference Waring, Rocca, Petersen, O'Brien, Tangalos and Kokmen29 Similarly, for observational studies and retrospective analysis, some participants could have changed oestrogen type or administration route, which leads to less accurate results.Reference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8,Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16,Reference Kim, Soto, Branigan, Rodgers and Brinton20
Selection biases are bound to occur in case–control studies. For example, data for progestin use were unobtainable, therefore the effect between unopposed oestrogen and oestrogen plus progestin was not determined.Reference Henderson, Benke, Green, Cupples and Farrer9
Only studies that were published in English were included, leading to a language bias in our results and findings. Positive findings are also more likely to be published in English as it is the predominant language in research.Reference Maki and Dumas37 Most studies are conducted in countries with high White populations, and this can introduce some ethnic bias as other ethnicities had little involvement in the studies.
Although many studies have taken place, the results remain inconsistent and insufficient to justify oestrogen replacement therapy's role in reducing risk of Alzheimer's disease. A major limitation to this systematic review is the few RCTs available that matched our inclusion criteria.
To conclude, our hypothesis stating ‘oestrogen replacement therapy is likely to reduce the risk of Alzheimer's disease in post-menopausal women’ is supported to a certain extent, although further studies are required. To support this theory, we recommend further research on longer oestrogen replacement therapy duration, because studies have yet to find an effective time period, except for a single study with a trial duration of 4 years.Reference Kantarci, Lowe, Lesnick, Tosakulwong, Bailey and Fields7 Another consideration could be possible interaction between oestrogen and other treatment modalities, as some studies found a potential role of non-steroidal anti-inflammatory drugs.Reference Zandi, Carlson, Plassman, Welsh-Bohmer, Mayer and Steffens8
In addition to new findings of oestrogen involvement in gene modulation, further investigation is encouraged with larger sample sizes, to further establish association of oestrogen replacement therapy with the presence of different numbers of ɛ4 alleles.Reference Tang, Jacobs, Stern, Marder, Schofield and Gurland16
In women who are predisposed to Alzheimer's disease, initiating a pathway of preventive oestrogen replacement therapy or HRT administration can be considered. Depending on whether the patient has had a hysterectomy, patients without a uterus should undertake APOE ɛ4 genetic testing. If the use of preventive measures are proven by future studies, the presence of ε4 alleles in the APOE ɛ4 genetic testing results should encourage immediate initiation of oestrogen replacement therapy. For patients with an intact uterus, clinical judgement is heavily required to determine whether benefits of APOE ɛ4 genetic testing and HRT administration outweigh the risks. However, more research is required in this area, and the conclusions we have drawn here are only from the studies in this paper, which we acknowledge are insufficient to fuel clinical decisions.
Supplementary material
Supplementary material is available online at https://doi.org/10.1192/bjo.2023.579
Data availability
Data availability is not applicable to this article as no new data were created or analysed in this study.
Acknowledgements
We would like to acknowledge Dr Christopher Haxton (Goldenhill Mental Health Resource Centre, Clydebank, Scotland) for the advice and guidance he provided throughout the study.
Author contributions
G.W., E.L., Q.Y.L. and H.R. were all equally involved in writing the protocol, performing initial searches of databases, performing data extraction and quality assessment of included studies and writing up of the manuscript.
Funding
This research received no specific grant from any funding agency, commercial or not-for-profit sectors.
Declaration of interest
None.
Open access
eLetters
No eLetters have been published for this article.