Long-term effects of 4 years of menopausal hormone therapy on white matter integrity

Abstract

Objectives:

To assess the long-term effects of 4 years of menopausal hormone therapy (mHT) on the brain’s white matter architecture in women who initiated mHT within 3 years of menopause onset.

Methods:

The Kronos Early Estrogen Prevention Study (KEEPS) was a multicenter, double-blind, randomized, placebo-controlled 4-year mHT trial with treatment arms of oral conjugated equine estrogens (oCEE), transdermal 17β-estradiol (tE2), and placebo in recently postmenopausal women. KEEPS Continuation was an observational follow-up of KEEPS participants. White matter integrity was evaluated in KEEPS Continuation participants 10 years after KEEPS completion using white matter hyperintensity volume, diffusion magnetic resonance imaging (dMRI) techniques, and cerebral infarcts. Linear regression models were fitted for each brain region to evaluate if there were differences in white matter between KEEPS treatment arms.

Results:

There was no evidence to suggest the long-term effects of 4 years of mHT on brain white matter in KEEPS Continuation participants [n=266, mean age 67 (58-73)]. No differences in dMRI metrics were found in each of the treatment arms (oCEE n=70; tE2 n=79) when compared to placebo (n=94), following a false discovery rate adjustment for multiple comparisons. There were no statistically significant differences in white matter hyperintensity volume or infarct occurrence when comparing each of the treatment arms to placebo.

Conclusions:

We found no evidence of the long-term effect of 4-year mHT on white matter integrity when compared to placebo, consistent with emerging evidence of the safety of short-term use of mHT in recently postmenopausal women.

Ovarian hormones have been implicated as modulators of cognitive health in women. Women who experience premature or early menopause, including primary ovarian insufficiency, have been shown to be at greater risk for cognitive impairment and dementia of any type than their peers.^1-5 Although estrogen has been proposed as a neuroprotective hormone in women, the Women’s Health Initiative (WHI) Memory Study (WHIMS) demonstrated that oral conjugated estrogens plus medroxyprogesterone acetate (MPA) were linked with an increased risk in dementia and negative impact on global cognitive function, and oCEE alone was linked with adverse effect on global cognitive function in women who were 65 or older.⁶ Previous studies have posited the existence of a critical age window in which menopausal hormone therapy provides cognitive benefits in perimenopausal or early postmenopausal women.^3,5,6 Large, randomized trials such as the Kronos Early Estrogen Prevention Study (KEEPS), Women’s Health Initiative Memory Study of Younger Women (WHIMS-Y), Cognitive Complaints in Early Menopause (COGENT), and the Early versus Late Intervention Trial with Estradiol (ELITE) tested the clinical time hypothesis using different formulations and doses of estrogens with or without synthetic or natural progesterone.^6-10 The randomized clinical trials demonstrated no significant cognitive harm or benefits of mHT when given within 5 years after menopause.⁸ Contributing to the mixed picture, additional case-control studies in Denmark and Finland demonstrated an association between hormone therapy and dementia regardless of time past menopause.^11,12 However, the long-term effect of short-term mHT on brain health remains to be investigated.

Noninvasive magnetic resonance imaging (MRI) is a powerful neuroimaging tool to detect early microstructural changes in the brain and predict future cognitive alterations.^13-15 Diffusion tensor imaging (DTI), a widely used diffusion MRI (dMRI) method, quantifies water diffusion in brain tissue yielding measures such as fractional anisotropy (FA) and mean diffusivity (MD). These metrics are commonly employed to investigate microstructural changes in the brain’s white matter tracts, which are formed by myelinated axonal fibers. A decrease in FA can indicate a loss of myelin integrity. An increase in MD reflects increased water mobility in tissue, typically due to the loss of structural barriers that restrict water diffusion, like myelinated fibers.

Recent developments in dMRI have introduced more sophisticated multishell models, such as Neurite Orientation Dispersion and Density Imaging (NODDI). NODDI provides biologically relevant parameters that offer more specific insight into neuronal integrity and organization: the neurite density index (NDI, the density of axons and dendrites), orientation dispersion index (ODI, the directional spread of the neurites), and isotropic volume fraction (ISOVF, the proportion of freely diffusing water molecules). NODDI provides several advantages over traditional DTI metrics such as FA and MD by providing greater characterization of microstructural features of brain tissue. A lower NDI has been shown to reflect degeneration in white matter axonal tracts. Changes to ODI can be more challenging to interpret; a higher ODI can reflect dendritic degeneration while a lower ODI can suggest decreased axonal organization.¹³

Alterations in dMRI metrics of white matter, such as a decrease in FA and an increase in MD, have been found in individuals with mild cognitive impairment and Alzheimer’s dementia before the manifestations of overt cognitive impairment.^16-18 This suggests that white matter changes modeled through dMRI may have predictive value for early loss of white matter integrity, implying the risk of cognitive impairment. In addition to microstructural changes detectable by dMRI, macrostructural alterations in white matter are also associated with a greater risk of cognitive impairment. WMHs, visible as bright signals on FLAIR MRI scans, are often associated with small vessel ischemic disease in the brain and represent another important predictor for subsequent cognitive decline.^19,20

Investigation of white matter architecture may provide further insight into the influence of short-term mHT on cognitive health in recently postmenopausal women. KEEPS, a randomized, double-blind, multisite, placebo-controlled 4-year clinical trial of mHT (oral conjugated equine estrogens, oCEE; and transdermal 17β-estradiol, tE2), provides an ideal cohort to investigate the long-term impact of 4-year mHT trial on brain white matter architecture when initiated early after menopause given the mitigation of many confounding factors through rigorous screening for cardiovascular risk factors. In an ancillary study of KEEPS, we demonstrated that tE2 was associated with better preservation of prefrontal cortex volume (a gray matter region) compared to placebo 7 years post-randomization (ie, 3 y after the end of the 4-year mHT trial).²¹ Although the previous study investigates gray matter over time, we hypothesized that the neuroprotective effect of tE2 (over placebo) would reflect in white matter integrity bridging the same regions 14 years after randomization in the KEEPS study.

In this study, we investigated the long-term impact of two common formulations of mHT versus placebo on brain white matter health ~14 years post-randomization. Specifically, we examined white matter hyperintensities, the presence of infarcts, and regional DTI and NODDI parameters in brain white matter tracts.

METHODS

Study participants

The original KEEPS trial (2005-2012) enrolled 727 women between 42 and 58 years of age, within 6-36 months of their last menses and with low cardiovascular risk. KEEPS participants all had intact uteruses at enrollment and had undergone spontaneous menopause. Full inclusion and exclusion criteria are described elsewhere.²² KEEPS participants were randomized to one of three treatment arms: oCEE (Premarin, 0.45 mg/d; Pfizer Pharmaceuticals), tE2 (Climara skin patch, 50 µg/d; Bayer HealthCare Pharmaceuticals Inc.), or placebo pills or patch. In addition, participants randomized to either mHT arm received micronized progesterone (Prometrium, 200 mg/d; Abbott Laboratories) for 12 days at the beginning of the month, while the placebo group received placebo progesterone. KEEPS enrollment occurred from August 2005 to July 2008, and participants were maintained on mHT or placebo for 4 years with a staggered trial ending. Briefly, women were excluded if they had a history of clinically defined cardiovascular disease, including myocardial infarction, angina, congestive heart failure, stroke, transient ischemic attack, or thromboembolic disease, uncontrolled hypertension (systolic BP>150 mm Hg or diastolic BP>95 mm Hg), smoking >10 cigarettes daily, a body mass index (BMI) >35 kg/m², diabetes (fasting glucose >126 mg/dL), dyslipidemia (total cholesterol >240 mg/dL), or a coronary artery calcium (CAC) score of 50 Agatston units or greater.

KEEPS Continuation was an observational cohort study that investigated the long-term impact of mHT ~14 years after randomization in the original KEEPS trial. The recruitment occurred between May 2017 and June 2022. Further explanation of participant enrollment in KEEPS Continuation is described elsewhere.²³ Of the 299 participants of KEEPS Continuation at seven sites (Banner Alzheimer’s Institute, Brigham Women’s Hospital, Columbia University, Mayo Clinic Rochester, University of California San Francisco, University of Utah, and Yale University; participants from the Albert Einstein College of Medicine/Montefiore Medical Center site in New York were enrolled at Columbia University), 266 underwent MRI imaging. Quality control was performed on MRI scans for protocol and scan parameters. Twenty-three of the MRI scans were either non-multishell (n=13) or failed dMRI protocol or quality control (n=10), resulting in a dMRI subset of 243 (oCEE, n=70; tE2, n=79; and placebo, n=94) (Fig. 1). Meanwhile, nine scans were not used in the WMH subset due to failure of FLAIR (Fluid attenuated inversion recovery, an MRI sequence often used for detection of superficial brain lesions) protocol compliance, WMH mask under or overestimating, or data errors (did not complete FLAIR and data not appropriately transferred to the central lab) resulting in a WMH subset of 257 (oCEE, n=74; tE2, n=85; and placebo, n=98) (Fig. 1). Infarct analysis also required FLAIR sequence; therefore, the infarct analyses were run for WMH subset participants.

FIG. 1.

Flow chart of KEEPS Continuation participants and subsets for the present study. DTI, diffusion tensor imaging; mHT, menopausal hormone therapy; MRI, magnetic resonance imaging of the brain; oCEE, oral conjugated equine estrogens; tE2, transdermal 17β-estradiol; WMH, white matter hyperintensity.

Participant demographics at KEEPS Continuation including age, APOE ε4 carrier status, smoking status (current vs. not current), education, body mass index (BMI), waist/hip ratio, systolic and diastolic blood pressures, lipid profiles, insulin, glucose, diabetes medication use, and type and duration of mHT use after KEEPS were available from the KEEPS Continuation database. APOE ε4 carrier status was determined through genotyping. Date of birth, current smoking status, years of education, and medication use were self-reported on a questionnaire. BMI, waist/hip ratio, and systolic and diastolic blood pressures were measured at KEEPS Continuation visits. Fasting glucose, insulin, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, Non-high-density lipoprotein cholesterol, total cholesterol, and triglycerides were evaluated with standard protocol venipuncture at Mayo Clinic Laboratories. Homeostasis Model Assessment of Insulin Resistance was calculated with glucose and insulin levels. Some participants in KEEPS Continuation (n=38) continued, initiated, or changed systemic mHT after cessation of randomized intervention of KEEPS.

Brain magnetic resonance imaging examinations

The MRI examinations were performed on Siemens Prisma, Skyra, Vida, and GE Discovery MR750 scanners across seven sites in the KEEPS Continuation. MRI protocol was harmonized across the participating sites using the standards set by the Alzheimer’s Disease Neuroimaging Initiative (ADNI).²⁴ All participants underwent a 3T head MRI protocol that included a magnetization prepared rapid gradient echo (MPRAGE) T1-weighted sequence (TR/TE/TI=2300/2.98/900 ms, flip angle 9 degrees, and 1.0 mm isotropic resolution), a 3D T2-FLAIR sequence (TR/TE/TI=4800/441/1550 ms, variable flip angle, resolution 1.0×1.0×1.2 mm, 3x GRAPPA acceleration, fat suppression on), and a dMRI. The dMRIs were spin-echo single shot Echo Planar Imaging with 2.0 mm isotropic voxels. Each Siemens dMRI had 13 b=0, 6 b=500, 48 b=1000, and 60 b=2000 s/mm² volumes, with the diffusion gradients in each shell evenly spread using an electrostatic repulsion scheme modified to distribute them over whole spheres instead of hemispheres. The GE dMRIs had 6 b=0 volumes and 48 b=1,000 s/mm² diffusion-weighted directions evenly spread over the whole sphere using a single shell scheme.^25,26 Quality control procedures for protocol compliance and image quality were run centrally at the Mayo Clinic site. Quality control measures included visual inspection and correction of head motion and current distortion.

Diffusion magnetic resonance imaging preprocessing

We used an in-house pipeline to preprocess the diffusion data sets, which were visually inspected by trained analysts blinded to the original treatment group. An intracranial mask was created for the diffusion MRI scan and noise was estimated and removed using random matrix theory.^27,28 Head motion and eddy current distortion were corrected using FSL’s eddy_cuda,²⁹ followed by correction of Gibbs ringing³⁰ and Rician bias.³¹ Diffusion tensors were fitted using a nonlinear least-squares fitting algorithm implemented in DIPY³² to generate FA and MD images.

The NODDI model was fitted using the Accelerated Microstructure Imaging via Convex Optimization implementation in Python,³³ producing voxel-wise maps of ODI, NDI, and ISOVF. We calculated a tissue-weighted NDI (tNDI) by adjusting the original NDI measurement with a scaling factor based on (1-ISOVF), which accounts for the presence of free water, cerebrospinal fluid, and other extracellular spaces. tNDI mixes NDI and ISOVF, but unlike NDI it does not suffer from a loss of precision when the tissue fraction is low, and unlike ISOVF, which is often 0, tNDI supports log-transformed statistical analysis. We present both NDI and tNDI to maintain the distinction between in-tissue microstructural changes and changes in the amount of free water. To generate regional measures of FA, MD, NDI, tNDI, and ODI for each bilateral white matter tract, we first registered the JHU “Eve” White Matter atlas³⁴ to participant native space using Advanced Normalization Tools—Symmetric Normalization (https://stnava.github.io/ANTs/).³⁵ Next, bilateral median values of FA, MD, tNDI, and ODI were computed, weighted by the size of each region of interest.

White matter hyperintensity analysis

WMH volumes were measured using an in-house, previously published, fully automated segmentation method.¹⁹ Briefly, T1-weighted MRIs were segmented using SPM12 with tissue priors and settings from the Mayo Clinic Adult Lifespan Template (MCALT). Corresponding 3D T2-FLAIR MRIs were co-registered using SPM12, and the T1-weighted segmentations were resampled to T2-FLAIR space. Voxels with significantly greater intensity than a sample of gray matter were considered candidate WMH locations, which were clustered using connected components, and some clusters were excluded as false positives based on size, intensity after blurring, or relative location in WM versus GM. Each automated segmentation was inspected by a trained image analyst to confirm successful quantification. Voxels associated with infarcts were removed and not considered as part of the WMH measurement. Total WMH volumes were measured as the sum of detected voxels, multiplied by the image’s voxel volume. It has been demonstrated that automated in-house algorithm to measure WMH volume is highly correlated with visual rating scale.³⁶

Infarct analysis

Cortical and subcortical infarcts were identified by trained analysts blinded to the treatment group and confirmed by radiologists. The full protocol has been defined in previous literature.³⁷ Briefly, infarcts were identified as hypointense lesions circumscribed by a hyperintense rim; hyperintensities associated with infarct pathology were not counted in WMH volume.

Statistical analysis

Characteristics of the participants were summarized using means and ranges for continuous data or counts and percentages for categorical data. Omnibus P values comparing the three groups (oCEE, tE2, and placebo) were calculated from a one-way analysis of variance for continuous variables or the Fisher exact test for categorical variables. If the omnibus tests were significant, then two pairwise comparisons of each treatment group compared to placebo were tested. Fisher exact tests were also performed comparing any differences between the three groups and the following variables: any infarcts indicator, count of infarcts, subcortical infarcts indicator, count of subcortical infarcts, cortical infarcts indicator, and count of cortical infarcts. We also assessed differences between the three groups and an ordinal infarct variable, zero, one, or two or more infarcts. All tests mentioned in this paragraph were performed for all KEEPS Continuation participants who had an MRI, the subset of those with dMRI data, and the subset of those with WMH data.

For each brain region, we fitted linear regression models using log-transformed WMH or dMRI values as outcomes. The predictor of interest was the KEEPS intervention group. We additionally adjusted for age, study site, and total intracranial volume (TIV) for WMH and for age for dMRI outcomes. Notably, instead of adjusting for a site for dMRI outcomes, ComBat was applied to attenuate the potential effects of the site. ComBat is a commonly used harmonization method in neuroimaging that estimates site effect or bias and removes unwanted variation from site while maintaining biological variability such as age.³⁸ For modeling, age was centered at the mean, TIV was centered at 1000 cm³ and log-transformed, and treatment was entered as a categorical variable with “placebo” set as the reference group. dMRI analysis results were adjusted for multiple comparisons using the false discovery rate (FDR) for the number of regions evaluated per outcome with the significance of Q<0.05.³⁹ Sensitivity analyses were performed by repeating all analyses, excluding participants who continued, initiated, or changed systemic mHT after KEEPS.

RESULTS

Participant characteristics

The characteristics of KEEPS Continuation participants with available MRIs (n=266) were compared across KEEPS randomized treatment groups. Demographics for the randomized treatment groups were similar except for diabetes medication use and oCEE use after cessation of randomized treatment (Table 1). The mean age at KEEPS Continuation was 67 (range 58-73 y). The percentages of participants who were APOE ε4 carriers were not statistically different among KEEPS randomization arms; 30% (n=20) of participants in the oCEE intervention arm, 32% (n=25) in the tE2 arm, and 18% (n=16) in the placebo arm were APOE ε4 carriers (P=0.08). There was a statistically significant difference between groups in diabetes medication use; the tE2 group (n=88) had significantly fewer participants who used diabetes medications compared to placebo (n=101; 1% compared to 8%, P=0.04). In addition, oCEE use after cessation of randomized treatment was different between treatment groups (4% of oCEE, 0% of both tE2 and placebo treatment arms, P=0.02), although pairwise P values were not significant. The difference in diabetes medication use in KEEPS Continuation participants has been previously noted and discussed in the literature,²³ while the impact of the difference in oCEE use after randomized treatment is further explored through sensitivity analyses performed by removing participants who continued to use mHT after the end of the KEEPS trial. Subsets of the 266 participants had MRI scans eligible for dMRI analysis (n=243) and FLAIR for WMH analysis (n=257). The KEEPS Continuation participant characteristics in the dMRI and WMH subsets were comparable among the treatment groups; however, no differences were found between any treatment arm with regards to diabetes medication use in the dMRI subset (P=0.08).

TABLE 1.

KEEPS Continuation participant characteristics at 14-year follow-up

		KEEPS randomization treatment group
	All (N=266)	oCEE (n=77)	tE2 (n=88)	Placebo (n=101)	P
Age at imaging	67 (58, 73)	67 (58, 72)	67 (59, 73)	67 (58, 73)	0.94
APOE ε4 carrier, n (%)	61 (26)	20 (30)	25 (32)	16 (18)	0.08
Education (y)	16 (12, 30)	16 (12, 20)	16 (12, 24)	16 (12, 30)	0.97
BMI (kg/m2)	26.5 (17.7, 50.8)	26.6 (18.4, 50.8)	26.2 (17.7, 36.0)	26.8 (18.3, 38.5)	0.74
Waist/hip ratio	0.85 (0.70, 1.23)	0.86 (0.70, 1.16)	0.85 (0.70, 0.97)	0.85 (0.70, 1.23)	0.73
Diabetes medication use, n (%)	10 (4)	1 (1)	1 (1)	8 (8)	0.03ab
Lipid-lowering medication use, n (%)	51 (19)	18 (23)	11 (12)	22 (22)	0.14
Anti-hypertension medication use, n (%)	70 (26)	19 (25)	20 (23)	31 (31)	0.44
Current smoker, n (%)	12 (5)	1 (1)	3 (3)	8 (8)	0.11
Systolic BP (mm Hg)	127 (86, 181)	126 (90, 165)	128 (95, 166)	128 (86, 181)	0.77
Diastolic BP (mm Hg)	76 (45, 106)	76 (50, 100)	76 (55, 103)	76 (45, 106)	0.98
Non-HDL (mg/dL)	139 (40, 277) [28%]	140 (65, 277) [28%]	141 (40, 207) [26%]	137 (46, 251) [28%]	0.80
Total cholesterol (mg/dL)	207 (104, 324) [20%]	207 (124, 324) [20%]	208 (113, 303) [19%]	206 (104, 313) [20%]	0.94
HDL (mg/dL)	67 (33, 122) [26%]	66 (36, 109) [26%]	67 (33, 122) [26%]	68 (37, 113) [25%]	0.73
LDL (mg/dL)	118 (15, 233) [30%]	118 (55, 233) [31%]	119 (25, 189) [30%]	117 (15, 215) [30%]	0.93
Triglycerides (mg/dL)	108 (39, 309) [47%]	110 (42, 291) [47%]	111 (40, 252) [41%]	103 (39, 309) [51%]	0.45
Glucose (mg/dL)	93 (74, 153) [11%]	93 (74, 153) [11%]	93 (77, 146) [11%]	93 (75, 147) [12%]	>0.99
Insulin (uIU/mL)	4.3 (0.6, 47.3) [89%]	4.6 (1.1, 47.3) [118%]	4.1 (0.6, 13.8) [71%]	4.2 (0.7, 18.0) [69%]	0.66
HOMA-IR	1.0 (0.1, 17.9) [123%]	1.2 (0.2, 17.9) [176%]	1.0 (0.1, 3.0) [76%]	1.0 (0.1, 4.4) [79%]	0.53 c
Hormone therapy after KEEPs, n (%)	55 (21)	22 (29)	16 (18)	17 (17)	0.14 d
Transdermal estradiol, n (%)	16 (6)	5 (6)	8 (9)	3 (3)	0.20
Transdermal estrogen, n (%)	1 (0.4)	0	1 (1)	0	0.62
Transdermal compounded/bio-identical estrogen, n (%)	4 (2)	1 (1)	2 (2)	1 (1)	0.83
Oral estradiol, n (%)	9 (3)	5 (6)	1 (1)	3 (3)	0.18
Oral compounded/bio-identical estrogen, n (%)	4 (2)	1 (1)	2 (2)	1 (1)	0.83
Oral CEE, n (%)	3 (1)	3 (4)	0	0	0.02 e
Oral conjugated estrogen, n (%)	1 (0.4)	0	0	1 (1)	>0.99
Vaginal estradiol, n (%) f	16 (6)	7 (9)	2 (2)	7 (7)	0.15
Vaginal compounded/bio-identical estrogen, n (%) e	1 (0.4)	0	0	1 (1)	>0.99

BMI, body mass index (weight in kilograms divided by the square of height in meters); BP, blood pressure; CEE, conjugated equine estrogen; Cholesterol total, sum of blood’s cholesterol content (LDL + HDL+ triglycerides); DBP, diastolic blood pressure; HDL, high-density lipoprotein cholesterol; HOMA-IR, Homeostasis Model Assessment of Insulin Resistance; KEEPS, Kronos Early Estrogen Prevention Study; Non-HDL, measured by subtracting HDL from total cholesterol; waist/hip (waist-to-hip ratio), waist circumference divided by hip circumference.

Continuous demographic data are presented as mean (range), lab values are shown as mean (range) [CV%], and categorical variables are presented as frequency and percentage [n(%)]. This table includes all available participants (n=266); however, there are missing characteristics: APOE ε4, n=30; education, n=1; SBP, DBP, n=1; non-HDL, total cholesterol, HDL, LDL, triglycerides, n=6; glucose n=16; insulin, n=18. Overall P values are from the analysis of variance or the Fisher Exact Test where appropriate.

^a tE2 is statistically different from placebo (P=0.04).

^b Within the dMRI subgroup, the difference between tE2 and placebo was no longer significant (P=0.08).

^c P value is from log-transformed HOMA-IR.

^d P value is from Yes/No (not distribution of hormone types).

^e Pairwise P value between groups is not significant.

^f Vaginal estradiol and vaginal compounded/bio-identical estrogen are not considered systemic hormone types.

Diffusion magnetic resonance imaging

dMRI findings are reported as log-unit differences relative to placebo (ie, treatment minus placebo). Five DTI outcome variables were investigated in 39 different white matter regions and tracks (Fig. 2). There were no statistically significant differences in FA, MD, NDI, or tNDI among treatment groups in any of the white matter regions and tracks after FDR correction. The interaction of APOE ε4 with the treatment group was explored for dMRI values. After FDR correction, there were no significant interactions between either intervention arm and APOE ε4. There were differences in FA and ODI before FDR correction, which are further characterized here.

FIG. 2.

dMRI regression results from best model log-transformed values relative to placebo in white matter regions before FDR correction for multiple comparisons. FA, fractional anisotropy; MD, mean diffusivity; ODI, orientation dispersion index; NDI, neurite density index; tNDI: tissue-weighted neurite density index. Higher FA, NDI, and tNDI and lower MD and ODI correspond to increased axonal organization. Brain region abbreviations are as follows: ACR, anterior corona radiata; ALIC, anterior limb of internal capsule; AWM, angular white matter; BCC, body of corpus callosum; CGC, cingulum cingulate gyrus; CGH, cingulum hippocampus; CST, corticospinal tract; EC, external capsule; ENT, entorhinal area; Fx, fornix; Fx_ST, fornix stria terminalis; GCC, genu of corpus callosum; IFO, inferior fronto-occipital fasciculus; IFWM, inferior frontal white matter; IOWM, inferior occipital white matter; ITWM, inferior temporal white matter; LFOWM, lateral fronto-orbital white matter; MFOWM, middle fronto-orbital white matter; MFWM, middle frontal white matter; MOWM, middle occipital white matter; MTWM, middle temporal white matter; PCR, posterior corona radiata; PLIC, posterior limb of the internal capsule; PoCWM, postcentral white matter; PrCWM, precentral white matter; PTR, posterior thalamic radiation; RLIC, retrolenticular part of the internal capsule; RWM, rectus white matter; SCC, splenium of corpus callosum; SCR, superior corona radiata; SFO, superior fronto-occipital fasciculus; SFWM, superior frontal white matter; SLF, superior longitudinal fasciculus; SMWM, supramarginal white matter; SOWM, superior occipital white matter; SPWM, superior parietal white matter; SS, sagittal stratum; STWM, superior temporal white matter; UNC, uncinate fasciculus.

Fractional anisotropy

Within the corticospinal tract, the oCEE and tE2 groups demonstrated higher FA relative to placebo. A higher FA in WM relative to placebo suggests the level of axonal organization is higher in mHT groups compared to placebo. Neither the FA differences in tE2 nor the oCEE groups relative to placebo persisted after correcting for multiple comparisons (Q=0.08, Q=0.86). The tE2 group additionally demonstrated lower FA from placebo within the inferior temporal white matter and lateral fronto-orbital white matter (Q=0.15, Q=0.58).

Orientation dispersion index

Higher ODI suggests decreased organization of neuronal tracts, indicative of degeneration. Within the inferior temporal white matter and lateral fronto-orbital white matter, tE2 ODI values were higher than the placebo group (Q=0.21; Q=0.30). The oCEE group demonstrated lower ODI in the angular white matter and body of the corpus callosum than placebo (Q=0.61, Q=0.61). None of these findings survived a correction for multiple comparisons.

White matter hyperintensity

There was no evidence of a difference in WMH between either treatment arm relative to placebo (oCEE, P=0.21; tE2, P=0.43) (Fig. 3).

FIG. 3.

Log-transformed white matter hyperintensity volume (95% CI) of treatment arms relative to placebo. oCEE, oral conjugated equine estrogens; tE2, transdermal 17β-estradiol; WMH, white matter hyperintensity

Interactions of WMH with APOE ε4 were investigated to evaluate the impact of possible modifying variables. There was no significant interaction between APOE ε4 carrier status and the treatment arm.

Infarcts

Thirty-five total participants (14%) had at least one infarct, the majority of which were subcortical (n=30) (Table 2). The percent of participants with at least one (≥1) subcortical infarct, the percent with at least one cortical infarct, and the percent of participants who were infarct positive (≥1 total infarct) were not different between oCEE, tE2, and placebo arms (P=0.94, P=0.76, P=0.95) (Fig. 4A). Furthermore, we did not find a difference in the frequency of infarcts (zero, one, or two or more) among the oCEE, tE2, and placebo arms (P=0.90; Fig. 4B).

TABLE 2.

Distribution of infarcts in all WMH subset (n=257)

	All (n=257)	oCEE (n=74)	tE2 (n=85)	Placebo (n=98)	P
Infarct positive, n (%)	35 (14)	11 (15)	11 (13)	13 (13)	0.95
No. of infarct, n (%)					0.90
0	222 (86)	63 (85)	74 (87)	85 (87)
1	25 (10)	7 (9)	9 (11)	9 (9)
2+	10 (4)	4 (5)	2 (2)	4 (4)
Subcortical infarct positive, n (%)	30 (12)	9 (12)	9 (11)	12 (12)	0.94
No. of subcortical infarct, n (%)					0.99
0	227 (88)	65 (88)	76 (89)	86 (88)
1	22 (9)	6 (8)	7 (8)	9 (9)
2+	8 (3)	3 (4)	2 (2)	3 (3)
Cortical infarct positive, n (%)	9 (4)	2 (3)	4 (5)	3 (3)	0.76
No. of cortical infarct, n (%)					0.57
0	248 (96)	72 (97)	81 (95)	95 (97)
1	7 (3)	1 (1)	4 (5)	2 (2)
2+	2 (1)	1 (1)	0 (0)	1 (1)

oCEE, oral conjugated equine estrogens; tE2, transdermal 17β-estradiol.

Data shown are n (% of treatment group). Groupwise P values comparing treatment to placebo are from the Fisher exact test.

FIG. 4.

(A) Percentage of participants in each treatment arm with ≥1 infarct (subdivided into total infarcts, subcortical, and cortical infarcts) in each treatment arm. (B) Percentages of participants in each treatment arm who have ≥1 infarct (are infarct positive), have one infarct, and have more than two or more infarcts. oCEE, oral conjugated equine estrogens; tE2, transdermal 17β-estradiol.

Sensitivity analysis

Sensitivity analyses were performed by removing participants who used systemic mHT (including transdermal or oral estradiol, estrogen, compounded bio-identical estrogen, and conjugated estrogen) after the cessation of randomized treatment from the analysis. Baseline demographics after exclusion from the main sample (266-38 results in n=228) were not different between treatment groups.

With regard to dMRI results (243-31 results in n=212), the findings of the sensitivity analyses were similar to the main analysis. As was the case in the main analyses, none of the results in the sensitivity analysis persisted after correction for multiple comparisons. Again, there were no interactions between APOE ε4 and either treatment group after a correction for multiple comparisons was applied.

The WMH results were unchanged in the sensitivity analysis (257-37 results in n=220). Similarly, no differences were found when investigating the relationship between the three treatment arms and infarcts, subcortical infarcts, or cortical infarcts, unchanged from the main analysis.

DISCUSSION

The present study did not find any evidence of long-term benefit or risk on white matter integrity and cerebrovascular lesions in women exposed to 4 years of mHTs (oCEE or tE2 and cyclic progesterone) compared to placebo after correcting for multiple comparisons. Brain white matter integrity was investigated using WMH volume, dMRI metrics, and the total number of infarcts in KEEPS Continuation participants. Initial dMRI variable comparisons depicted a mixed picture of greater and lesser white matter integrity in various white matter regions, although there were no consistent trends among the five dMRI metrics (FA, MD, NDI, ODI, and tNDI) for mHT users compared to placebo in any of the 39 investigated regions, nor did any of the trends persist after the application of a correction for multiple comparisons. Correction for multiple comparisons is necessary for a study such as this when data from many white matter regions and tracts in the brain are examined. Neither WMH volume nor number of infarcts differed significantly between treatment groups and placebo. These results suggest that short-term (4 y) use of combination mHT early in menopause does not appear to have long-lasting effects on white matter integrity, WMH volume, or cerebral infarct occurrence when compared to placebo.

Previous research has yielded mixed results on the long-term neurological and vascular safety of mHT. The WHI hormone trial found that women aged 50-79 years with prior hysterectomy had a greater risk of incident stroke and dementia with estrogen-only hormone therapy compared to placebo.⁴⁰ The WHIMS trial determined that both oCEE with MPA and oCEE alone had deleterious effects on global cognition in women aged 65 years and older.^41-43 Furthermore, both treatment groups demonstrated an increased risk of stroke during treatment but not in the follow-up period.⁴⁴ Subsequent studies, including the ELITE trial, demonstrated a beneficial effect of early initiation of HT on carotid intima-media thickness, a marker of cerebral atherosclerosis,⁴⁵ but neither KEEPS Continuation studies nor ELITE detected cognitive differences when comparing HT and placebo.^7,9 KEEPS Continuation has thus far demonstrated no adverse effects nor benefits of short-term (4 years) mHT on metabolic or cardiovascular outcomes 10 years after the end of a clinical trial.²³

Within the course of the KEEPS trial, neurological health has been evaluated through total brain volume, WMH, and cognitive function. In an ancillary MRI study, MRI and cognition tests were performed at baseline, 18, 36, 48, and 84 months after randomization to evaluate changes in brain structure and cognitive health of KEEPS participants at the Mayo Clinic site (n=75). There was no difference in the rate of change in global cognitive function between placebo and either treatment group. In the oCEE group, there was a greater increase in WMH volume compared to placebo observed at 4 years after randomization that normalized at year 7.^21,46

The present findings suggest that the increase in WMH seen 3 years after the end of the KEEPS trial normalized within 10 years after therapy cessation without long-term sequelae, including stroke or changes in white matter integrity. Moreover, there was no evidence of lower WMH in the treatment groups compared to placebo. Increased WMH volume has been associated with the severity of physiological vasomotor symptoms in menopausal women.⁴⁷ Although mHT did decrease subjective vasomotor symptom severity among participants in the KEEPS trial,⁴⁸ there was no beneficial or harmful impact of 4 years of mHT on WMH burden.⁴⁸ Our current data demonstrate no long-term effects, neither positive nor negative, of prior exposure to 4 years of mHT on white matter architecture in this cohort of postmenopausal women.

The KEEPS Continuation study has many strengths, including the long-term follow-up of participants originally randomized to treatment and placebo groups in KEEPS. The length of follow-up of ~14 years after randomization contributes to the data from the original KEEPS trial and other research with shorter durations of follow-up.^7,22,45 In addition, ongoing self-reported HT use after completion of the KEEPS trial allowed us to conduct a sensitivity analysis, which corroborated the main group findings of no effect and enhanced the reliability of our conclusions. A key methodological strength lies in our comprehensive approach to evaluating white matter microstructure using both DTI/NODDI and WMH, allowing for further characterization of white matter microstructural changes. Both WMH and DTI/NODDI analyses have been used as early indicators leading to dementia and neurological events such as stroke.^14,15,20 Using both in combination allowed the study to investigate white matter tract health with multiple measures, increasing insight into possible mechanisms of change in cognition previously noted in women who used HT.

Our findings should be considered within the context of several limitations. Only 37% of original KEEPS participants opted to participate in the KEEPS Continuation study and underwent MRI imaging (n=266). Subsets of the 266-person cohort were used in dMRI and WMH analyses due to failure of quality control or protocol, leading to a further decrease in sample size and power. Reasons for the decreased participation at KEEPS’ Continuation could include recruitment during the COVID-19 pandemic, when the public were increasingly uneasy about presenting to health care settings for testing, potentially skewing our participant pool toward those more willing or able to participate despite the circumstances. Furthermore, the use of differing MRI scanners across the multiple sites of the study may affect results despite the use of ComBat to remove the variation.

Prior analysis of the KEEPS cohort demonstrated that KEEPS Continuation participants had significantly lower systolic and diastolic blood pressure at the beginning of the original KEEPS study than members of the KEEPS cohort who did not participate in KEEPS Continuation.²³ The differences in blood pressure may impact the analysis of white matter given the association between white matter and blood pressure which, combined with the substantial attrition rate, further decreases the generalizability of the study.⁴⁹ Despite this, the present study maintained a sample size of at least n=243 for each analysis, and the treatment groups remained demographically similar.

CONCLUSIONS

The present findings are consistent with emerging research from the KEEPS Continuation trial demonstrating a lack of long-term effects, either adverse or beneficial, of a 4-year exposure to low-dose menopausal hormone therapy started shortly after menopause onset. Our study did not demonstrate any significant effects of short-term (4 years) mHT on measures of white matter integrity, although it is limited by statistical power.