Inside the KEEPS Methodology: What Most Summaries Skip

By
Published Updated Last reviewed
Hormone therapy clinical care image for Inside the KEEPS Methodology: What Most Summaries Skip

At a glance

| Parameter | Detail | |---|---| | N | 727 (randomized) | | Intervention | Oral CEE 0.45 mg/d or transdermal 17β-estradiol 50 µg/d, both with cyclic oral micronized progesterone 200 mg/d × 12 d/mo | | Comparator | Matching placebo pills + matching placebo patches | | Duration | 48 months | | Primary endpoint | Rate of change in carotid artery intima-media thickness (CIMT) | | Key result | No significant difference in CIMT progression between either active arm and placebo; secondary benefits in mood, vasomotor symptoms, and sexual function |

Why KEEPS Exists: The Post-WHI Problem

After the Women's Health Initiative (WHI) reported increased cardiovascular events with HRT in 2002, prescribing rates collapsed. But the average WHI participant was 63 years old, a full decade past menopause. Clinicians and researchers questioned whether starting hormones closer to menopause onset would produce different results. This became known as the "timing hypothesis."

KEEPS was purpose-built to test that hypothesis. Enrollment was restricted to women within 6 to 36 months of their final menstrual period, with a mean age of 52.7 years. That narrow window is the single most important design feature separating KEEPS from WHI and from earlier observational data.

Randomization and Blinding Architecture

KEEPS used a double-blind, placebo-controlled, randomized design across nine clinical centers in the United States. Participants were assigned 1:1:1 to one of three arms:

  1. Oral CEE (0.45 mg/day) + placebo patch
  2. Transdermal estradiol (50 µg/day patch) + placebo pill
  3. Placebo pill + placebo patch

The dual-dummy approach (every participant received both a pill and a patch) preserved blinding across the oral and transdermal arms. This is a design detail most summaries omit. Without it, participants and clinicians could infer assignment from the route of administration alone.

Cyclic oral micronized progesterone (200 mg for 12 days per 28-day cycle) was given to both active arms for endometrial protection. The placebo arm received matching progesterone placebos on the same schedule. This cyclic design meant some women experienced withdrawal bleeding, which could theoretically unblind participants. The investigators acknowledged this limitation but noted that bleeding patterns were variable enough across all arms that complete unblinding was unlikely for most participants.

Randomization was stratified by site using permuted blocks. Central allocation concealment was maintained through a coordinating center at the Kronos Longevity Research Institute.

Inclusion and Exclusion: A Deliberately Narrow Funnel

The enrollment criteria were designed to capture healthy, recently menopausal women:

Key inclusion criteria:

  • Age 42 to 58
  • Within 6 to 36 months of last menstrual period
  • FSH >35 mIU/mL (confirming menopausal status)
  • Good general health with no contraindications to HRT

Key exclusion criteria:

  • Body mass index >35 kg/m²
  • Smoking >10 cigarettes/day
  • Clinical cardiovascular disease (prior MI, stroke, VTE)
  • CIMT >50th percentile for age at screening (a critical filter, discussed below)
  • Uncontrolled hypertension, diabetes requiring medication, LDL cholesterol >190 mg/dL

The CIMT screening threshold deserves attention. Women whose baseline carotid wall thickness exceeded the 50th percentile for their age were excluded. This meant KEEPS enrolled a population with relatively healthy arteries. The rationale was sound: the trial aimed to test prevention, not treatment, of subclinical atherosclerosis. But it also means KEEPS results cannot be extrapolated to women who already have measurable vascular disease at the time of HRT initiation.

The narrow BMI cap (<35) further limits generalizability, given that the average BMI of postmenopausal women in the U.S. exceeds 30. These filters created a sample that was healthier, leaner, and whiter than the general menopausal population.

The Primary Endpoint: CIMT as a Surrogate

KEEPS chose carotid intima-media thickness, measured by B-mode ultrasound, as its primary endpoint. CIMT is a validated surrogate marker of subclinical atherosclerosis. Annual CIMT progression rates correlate with future cardiovascular events in large cohort studies, making it a plausible surrogate for a trial too small to capture hard clinical events like MI or stroke.

Measurements were taken at the common carotid artery and at the carotid bulb using a standardized protocol with central reading. Annual CIMT change (mm/year) was the primary analytic target. This matters because CIMT has different progression rates by location. The common carotid segment tends to progress more slowly and with less variability than the bulb or internal carotid. KEEPS analyzed both segments but designated the common carotid as the primary measure.

The choice of CIMT carries inherent limitations. A 2012 meta-analysis questioned whether CIMT adds predictive value beyond traditional risk factors for cardiovascular events (Den Ruijter et al., Lancet 2012). The surrogate endpoint was a pragmatic concession to sample size. With 727 women over 4 years, KEEPS was never powered to detect differences in actual cardiovascular events. This is the most common misunderstanding of the trial: a null CIMT result does not mean HRT has no cardiovascular effect, only that it did not measurably alter this specific surrogate in this specific population.

Comparator Logic: Why Two Active Arms?

Most HRT trials test one formulation. KEEPS tested two deliberately different delivery routes at deliberately lower doses than WHI used. This embedded a secondary hypothesis into the trial structure.

The KEEPS Comparator Framework:

| Design dimension | WHI (2002) | KEEPS (2014) | Why it matters | |---|---|---|---| | Oral CEE dose | 0.625 mg/d | 0.45 mg/d | Lower dose reduces hepatic first-pass effects, potentially fewer thrombotic signals | | Transdermal option | None | 50 µg/d patch | Bypasses liver entirely, different lipid and coagulation profile | | Progestogen | MPA 2.5 mg/d continuous | Micronized progesterone 200 mg cyclic | MPA has androgenic properties; micronized progesterone is considered more physiologic | | Mean participant age | 63 | 52.7 | Tests the timing hypothesis directly | | Years since menopause | ~12 | ~1.4 | Same |

This framework makes explicit what most KEEPS commentary leaves implicit. Every major variable that drew criticism in WHI was altered: dose, formulation, progestogen type, participant age, and time since menopause. The result is a trial that cannot be compared head-to-head with WHI on any single variable, because too many changed simultaneously.

From a design standpoint, the three-arm structure allowed both a treatment-vs-placebo comparison and an oral-vs-transdermal comparison. The transdermal arm was included because mechanistic data suggested that bypassing hepatic first-pass metabolism would produce a different thrombotic and lipid profile. Transdermal estradiol does not raise C-reactive protein or coagulation factors the way oral estrogens do (Canonico et al., Circulation 2007). KEEPS was positioned to detect whether these mechanistic differences translated into measurable differences in arterial wall outcomes.

Statistical Approach and Power

The primary analysis used a mixed-effects linear regression model, fitting CIMT as a function of time, treatment group, and a time-by-treatment interaction. The interaction term was the key inferential target: did the slope of CIMT change differ by treatment assignment?

KEEPS was powered to detect a 50% reduction in common carotid CIMT progression rate (from an assumed placebo rate of 0.01 mm/year) with 80% power at alpha = 0.05. The assumed progression rate was based on prior observational cohort data.

There are two issues with this power calculation worth noting:

  1. The observed placebo progression rate was lower than expected. Placebo-arm CIMT progressed at approximately 0.007 mm/year. With a slower underlying rate, the trial had less statistical power to detect a meaningful difference.
  2. The 50% reduction target was ambitious. Even potent lipid-lowering drugs like statins typically slow CIMT progression by 30 to 40% in trials of similar duration. Whether low-dose HRT would produce a 50% reduction was always uncertain.

The analysis was conducted on an intention-to-treat basis. Sensitivity analyses adjusted for adherence, but the primary results were intention-to-treat. Dropout rates were non-trivial at approximately 25% over 4 years, which is typical for a multi-year placebo-controlled trial requiring daily medication and patches but further dilutes the treatment signal in the ITT analysis.

What the Results Actually Show

The primary CIMT endpoint was null. Neither o-CEE nor t-E2 significantly altered the rate of common carotid CIMT progression compared to placebo.

| Outcome | o-CEE arm | t-E2 arm | Placebo | P-value (interaction) | |---|---|---|---|---| | CIMT progression (mm/yr, common carotid) | 0.007 | 0.007 | 0.007 | 0.89 (o-CEE), 0.85 (t-E2) | | Coronary artery calcium (Agatston score change) | +1.2 | +0.6 | +1.1 | NS for both | | Hot flash reduction | Significant | Significant |, | <0.05 | | Sexual function (FSFI) | Improved | Improved |, | <0.05 | | Mood/depression scores | Improved (o-CEE) | Trend |, | <0.05 (o-CEE) |

Coronary artery calcium (CAC) was a prespecified secondary endpoint. While the overall CAC comparison was not statistically significant, the o-CEE arm showed a trend toward higher progression in some subgroup analyses, though the clinical meaning of small CAC differences in women with low baseline scores remains debated.

The secondary quality-of-life outcomes are where KEEPS delivered clinically meaningful findings. Both hormone arms significantly reduced vasomotor symptoms and improved sexual function. The oral CEE arm also showed significant improvement in mood and depressive symptoms, consistent with prior smaller studies of estrogen's effects on serotonergic pathways.

Limitations the Authors Acknowledged

The original publication and companion papers noted several constraints:

  • Sample size. With 727 women, KEEPS was a mechanistic surrogate trial, not a clinical-events trial. It cannot speak to MI or stroke risk.
  • CIMT screening exclusion. Women with above-median baseline CIMT were excluded, limiting applicability to those with existing subclinical disease.
  • Duration. Four years may be too short to detect vascular remodeling differences that emerge over decades. The Early versus Late Intervention Trial with Estradiol (ELITE) used a 5-year design and found CIMT benefit only in the early-menopause subgroup, hinting that KEEPS may have been underpowered in time.
  • Homogeneous population. Over 90% of participants were white. Results may not generalize across racial and ethnic groups with different baseline cardiovascular risk profiles.
  • Withdrawal bleeding. Cyclic progesterone caused predictable bleeding patterns that could partially unblind participants.

How KEEPS Fits the Timing Hypothesis

KEEPS alone did not prove the timing hypothesis. But it filled a critical gap. Combined with ELITE and with reanalyses of WHI by age strata (which showed neutral or potentially beneficial trends in women aged 50 to 59; Rossouw et al., JAMA 2007), KEEPS contributed to the consensus reflected in the 2022 North American Menopause Society position statement that hormone therapy initiated near menopause onset in healthy women is not associated with increased cardiovascular risk and carries meaningful symptomatic benefits.

The trial's most durable contribution may not be the CIMT null result but its demonstration that low-dose, physiologic HRT formulations in recently menopausal women are safe over a 4-year horizon, with measurable improvements in quality of life.

Frequently asked questions

›
›
›
›
›
›
›
›
›
›

References

  1. Harman SM, Black DM, Naftolin F, et al. Arterial imaging and atherosclerosis in recently menopausal women: a randomized trial (KEEPS). Ann Intern Med. 2014;161(4):249-260. PubMed
  2. Hodis HN, Mack WJ, Henderson VW, et al. Vascular effects of early versus late postmenopausal treatment with estradiol (ELITE). N Engl J Med. 2016;374(13):1221-1231. PubMed
  3. Rossouw JE, Prentice RL, Manson JE, et al. Postmenopausal hormone therapy and risk of cardiovascular disease by age and years since menopause. JAMA. 2007;297(13):1465-1477. PubMed
  4. Canonico M, Oger E, Plu-Bureau G, et al. Hormone therapy and venous thromboembolism among postmenopausal women: impact of the route of estrogen administration and progestogens. Circulation. 2007;115(7):840-845. PubMed
  5. Den Ruijter HM, Peters SA, Anderson TJ, et al. Common carotid intima-media thickness measurements in cardiovascular risk prediction: a meta-analysis. JAMA. 2012;308(8):796-803. PubMed
  6. The 2022 Hormone Therapy Position Statement of The North American Menopause Society. Menopause. 2022;29(7):767-794. PubMed