Relevant foreign literature was searched by two independent researchers from databases including PubMed and Web of Science. The following keywords were used as search input: estrogen therapy, ERT, hormone therapy, hormone replacement therapy, Alzheimer's disease, AD, Parkinson's disease, and PD. There was no year restriction applied. Additional articles were selected from the reference section of certain publications. Only full-text journal articles with accessible data for analysis were included.

The initial search yielded 3,668 records from PubMed and 3,201 records from Web of Science. After the screening of titles and abstracts, 6,758 records were excluded because they were not related to our present subject. The remaining 111 articles were selected for full-text scrutiny. Ninety studies were excluded due to no usable data (n = 50), no control group (n = 21), meta-analyses studies (n = 13), or repeated analysis with some documents (n = 6). Therefore, a total of 21 studies with 1,266 patient cases and 3,845 control cases were included in this meta-analysis. A flowchart of the selection process was presented in Figure 1.

For each selected study, the following data were extracted: study design, number of participants, number of AD case, number of control case, participants' ages, method for collecting data on hormone use, follow-up time, year of publication, measure of effect, diagnostic criteria, classifications and frequencies of hormone therapy application (e.g., timing of use, duration of use, route of administration, formulation, or any available information) and model, or other covariates.

Age was provided in the form of the mean value, unless otherwise stated. Nearly all the studies included adjusted odds ratio (OR)/relative risk (RR)/hazard risk (HR) values because there were some differences in covariates among the studies.

The Comprehensive Meta-Analysis Version 2 software (Biostat, Englewood, NJ, USA) was used for all the statistical analyses. We grouped study findings on the basis of how hormone therapy was categorized (e.g., any vs. never used) and included a summary for measure of effect and 95% CIs in the tables. These summaries were calculated based on random-effects models which involved a weighting scheme.

Cochran Q test was applied to evaluate the statistical difference of heterogeneity across different studies. It was considered statistically significant when P < 0.05. I² index was used to determine the inconsistency across different studies to evaluate the impact of heterogeneity. We used 25, 50, and 75% of I² to define low, medium, and high levels of heterogeneity. The Egger's test was used to determine the significance of a statistical test for publication bias to assess the degree of asymmetry in the funnel plot.

Meta regression was conducted among factors that might lead to heterogeneity in order to identify the main factors. A predefined subgroup analysis was used to assess the impact of various factors in the study. The following subgroups were defined in the AD group: case >500 vs. case ≤500, case-control study vs. prospective cohort, publishing year ≤1995 vs. 1996–2005 vs. 2006–2019, women age ≤70 vs. 71–79 vs. age ≥80, measure of effect: OR vs. HR vs. RR, hormone therapy ascertainment by interview vs. questionnaires vs. prescription database vs. medical records, duration of the treatment <5 years vs. 5–10 years vs. treatment >10 years. It was considered as statistically significant if P < 0.05. Meanwhile, the change in I² was compared before and after the introduction of covariates into the regression model.

We found 111 potentially relevant articles. Among these articles, a total of 21 eligible studies were pooled together for analyses (Figure 1). The Newcastle–Ottawa Scale (NOS) scores of eligible articles were between 7 and 8, with an average of 7.57. Baseline characteristics of included studies are shown in Tables 1–3.

Stratification of the study: 13 were case-control studies, five were prospective cohort studies, and three were cross-sectional studies. Stratification of the location: 15 studies were conducted in America, three in Europe (one in the UK, one in Italy, and one in Netherlands), and two other countries (one in Canada, one in Australia). In the AD group, there were 13 studies in America, two in Europe (one in the UK and one in Netherlands), and one in other countries. Stratification of neurological disorders: five cases evaluated the impact of ERT on PD and 16 cases on AD. All studies were collected on the use of hormone therapy either by self-report (e.g., interview or questionnaire) at the start of the study, by electronic prescription database, or by medical records. Furthermore, all studies were included in this review except one reported using standard criteria to diagnose AD and dementia [e.g., National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA); Diagnostic and Statistical Manual of Mental Disorders, Third Edition, Revised (DSM-III-R); Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV); or Movement Disorder Society-Sponsored Revision Unified PD Rating Scale (MDS-UPDRS)].

We used random-effects meta-analysis to assess the association between ERT and neurological diseases. Our results showed that ERT decreased risks of developing AD (OR: 0.672; 95% CI: 0.581–0.779; P < 0.001) and PD (OR: 0.470; 95% CI: 0.368–0.600; P < 0.001) in patients compared with the control (Figure 2), suggesting that estrogen therapy had a greater impact on PD.

Further subgroup analyses by disease outcome, 16 studies also had small heterogeneity in AD (I² = 24.140; P = 0.181), but five studies of PD showed no heterogeneity (I² = 0.000; P = 0.558). Since the data of PD were not enough, we only performed meta-regression analysis on the AD group. Study design (P = 0.01) and effect measure (P = 0.03) might be the sources of heterogeneity in the AD group, but number of cases (P = 0.172), age (P = 0.986), publication year (P = 0.712), hormone therapy ascertainment (P = 0.494), and duration of the treatment (P = 0.217) had no moderating effects on the significant association between hormone replacement therapy (HRT) and AD incidence (P > 0.05 in these studies) (Table 4).

Furthermore, according to the results of subgroup analyses in the AD group, heterogeneity came from data measure. Two of the interaction terms of the predefined subgroups showed statistical significance: study design (P = 0.01) and measure of effect (P = 0.03). Estimated pooled differences among each subgroup are presented in Figure 3. For the stratified analyses among studies of AD, different measure of effects are the source of heterogeneity, but the root cause is different in effect design, suggesting that we should classify different research types before statistical analysis in meta-analysis. Forest plot displayed random-effects meta-analysis results for different effect designs in the AD subgroups (Figure 4). When only prospective cohort studies were included, we observed an increased effective size for AD studies (OR: 0.519; 95% CI: 0.413–0.653; P < 0.001), adding more proof that ERT is indeed beneficial for treating AD.

Sensitivity analysis demonstrated that none of the individual studies could induce statistical bias regarding the association between ERT and incidence of AD or PD, indicating that our findings were statistically reliable (Figure 5).

Funnel plots were used to assess publication biases. We did not find an obvious asymmetry of funnel plots in any of the comparisons, which suggested that our findings were unlikely to be impacted by severe publication biases (Figure 6).

ERT and neurodegenerative disease performance were related to factors such as age, country, socioeconomic status, and health status. It was not feasible to eliminate these factors from the epidemiological study. Such confounding factors can be the source of some analysis bias. First, there was a big challenge in selecting data since most studies did not report effects in a manner that allowed their results to be used for meta-analyses. There are differences in the determination of estrogen treatment results, treatment duration, or disease interval evaluation in the included articles. We added meta-regression analysis, which showed they were minimally relevant to the result in these subgroups. However, it would indeed become a source of limitation. Second, many of the observational studies showed a clear time-dependent pattern. The data included in this article have a large time span, thus the diagnostic criteria may change. To avoid this bias, we adopted NINCDS-ADRDA diagnostic criteria in AD studies and MDS-UPDRS in PD studies. Compared with other diagnostic criteria, they have been applied for more than 20 years since it was established. At the same time, we also used meta-regression to evaluate whether the diagnostic criteria have a trend of change with time, and the regression result is non-existent. The large controlled studies currently underway will hopefully address this time limit. Third, determining the history of hormone therapy use was a concern in studies that relied on self-report (e.g., interview or questionnaire). Pathological cognitive changes had been a big challenge to recall the memory of hormone therapy use. We included several studies with hospital records or multidisciplinary evaluations on top of patient's self-report to reduce the chance of recall bias.

Different researchers used appropriate study designs to study the relationship between ERT and AD and PD. Meta-regression results showed that the study design was indeed the heterogeneous source of meta-analysis (P = 0.01). Therefore, we classified the articles into case-control study and prospective cohort study according to different study designs and then re-conducted meta-analysis. When only prospective cohort studies were included, we observed an increased effective size for AD studies (OR: 0.519; 95% CI: 0.413–0.653; P < 0.001), adding more proof that ERT is indeed beneficial for treating AD (Figure 4). Therefore, we believe that this study design is more reasonable and effective in the epidemiological study of ERT and neurodegenerative diseases. We call on researchers to be more inclined to choose this research design in future epidemiological studies.

Some literature suggests that the role of ERT may depend on the age of menopause and the therapeutic intervention used. The time window of estrogen therapy is associated with the risk of onset and/or developing neurodegenerative disease, and early treatment performed 10 years after menopause can decrease the risk (Yaffe et al., 1998). Regression analysis for age (P = 0.98581 > 0.05) showed no statistical significance. According to the subgroup analysis among age ≤70 (I² = 33.396, P = 0.199), age 71–79 (I² = 38.876, P = 0.133), and age ≥80 (I² = 17.911, P = 0.296), there was no evidence indicating that age was associated with the risk of disease development.

What requires further investigation is the relationship between the route of administration of estrogen therapy and the risk of onset and/or developing neurodegenerative disease. Four studies differentiated the use by route of administration (oral vs. transdermal), as shown in Table 5. Meta-analysis results of the oral route were OR: 0.925, 95% CI: 0.618–1.385, and P = 0.707. Results of the transdermal drug delivery were OR: 0.975, 95% CI: 0.731–1.299, and P = 0.861. There was no statistical significance between the use of oral estrogens and transdermal estrogens. However, our sample size of only four studies might cause a bias in the result. There was a lack of evidence from large and randomized clinical trials that examine the efficacy and safety of alternative hormone therapy for the route of administration.

Through different regulatory mechanisms, estrogen affects the conduction of nerve signals and tissue changes in the brain. At the same time, genes associated with neurodegenerative diseases are also shown to be regulated by estrogen (Nilsson et al., 2011; Xing et al., 2013), and these results are in agreement with the results of our meta-analysis. It has been reported that estrogen decreases reactive oxygen leak and diffusion lipid peroxidation coupled with oxidative stress and endogenous oxidative damage by increasing electron transport chain complex IV and mitochondrial reactivity (Irwin et al., 2008). Brain-derived neurotrophic factor (BDNF) gene contains an estrogen response element (ERE), which confirms that ERβ affects the maturation and plasticity of synapses through the BDNF-TrkB signaling pathway (Zhao et al., 2011). We have shown that there is an important interaction between the apolipoprotein E (Apo E) gene and the risk of onset and/or developing AD (Liu et al., 2013). ERE presents on the Apo E gene, which can modify the expression of the Apo E gene in the cerebral cortex by 17β-estradiol (Struble, 2003). PD is a neurodegenerative disease caused by substantia nigra degeneration or loss of dopaminergic neurons. It has been found that estrogen can convert D2 DA receptors from a high affinity state to a low affinity state in monkeys with different dyskinesias. An important interaction between the brain renin-angiotensin system (RAS) and effects of 17β-estradiol in models of PD, the RAS enhances the progression of dopaminergic degeneration by intensifying neuroinflammation, and estrogen protects dopaminergic neurons by inhibition of RAS (Labandeira-Garcia et al., 2016). In the PD model, 17β-estradiol is a negative regulator of the RAS, which inhibits its function and reduces neuroinflammation and DA degeneration. Estrogen rapidly and directly acts on striatum and nucleus accumbens, via a G-protein-coupled external membrane receptor, to enhance DA releases and DA-mediated behaviors (Becker, 1999). At the same time, 17β-estradiol is found to inhibit 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP)-induced DA depletion under a dosing regimen (repeated daily administration) (Ramirez et al., 2003). At present, DA agonists are one of the main drugs for symptomatic treatment of PD. It is determined that the beneficial effects of estrogen on DA receptors can delay the progression of PD.

In conclusion, a meta-analysis was conducted with regard to the long-standing debate about whether ERT protects cognition and reduces the risk of neurodegenerative disease. First, different diseases were classified. In the case of AD, more research data were included, beneficial conclusions were thus obtained, which also verified the clinical observation data. Meta-analysis in the estrogen therapy and the risk of PD were first conducted, the results showed that estrogen therapy significantly reduced the risk of PD. These data can help with the development of new therapeutic ideas and preventative measures for future clinical application regarding the development AD and PD.

Some of the minor issues that have been experienced so far with estrogen use were addressed. The results of studies and meta-analysis indicated that estrogen therapy does have beneficial effects on neurodegenerative diseases such as AD and PD. Notably, neurodegenerative diseases are associated with internal energy and material metabolism disorders, which are not limited to reproductive hormones. According to the latest epidemiological studies, neurodegenerative diseases were closely related to diabetes and non-alcoholic fatty liver disease (NAFLD) (Szmuilowicz et al., 2009; Martins, 2014; Slopien et al., 2018; Venetsanaki and Polyzos, 2019). They may have a common pathogenic mechanism, which involves the production of Aβ protein, insulin resistance, and mitochondrial dysfunction (Martins, 2015, 2018). Detection of these endocrine markers that associate with metabolic syndrome would help with timely diagnosis of the disease in the early or presymptomatic phase. Future studies need to determine how the induction or inhibition of endocrinal targets could be used for predictable neuroprotection in neurodegenerative disease therapies.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.