Menopause in the All of Us Research Program: A Descriptive Summary of Electronic Health Record and Survey Response across Sociodemographic Characteristics

This study characterizes menopause-related data across survey, electronic health record, and genomic modalities in nearly 396,000 All of Us Research Program participants, revealing significant under-ascertainment in EHRs compared to self-reports and providing critical insights into age distributions and sociodemographic variations to guide future menopause research design.

The Gist

The Big Picture: A Massive Census of Women's Health

Imagine the All of Us Research Program as a giant, high-tech library containing the health stories of nearly half a million American women. The researchers wanted to know: "How well does this library actually record the story of menopause?"

Menopause is a major life event for women (usually happening around age 51), like a "season change" in the body that brings new health risks, such as heart issues or bone weakness. To study this, you need to know exactly who is in which "season" (before, during, or after menopause).

The team looked at two different ways the library tried to track this:

1. The Medical Record (EHR): The official, clinical notes doctors write down in the computer system.
2. The Survey: A questionnaire where women simply answer questions about their own bodies.

The Main Discovery: The "Silent" Medical Record

The most shocking finding was a massive gap between what doctors wrote and what women said.

• The Survey (The Truth-Teller): When women were asked, "Have your periods stopped permanently?" about 193,000 said "Yes."
• The Medical Record (The Quiet Observer): When the researchers looked at the doctors' official diagnosis codes, they only found about 28,000 women marked as menopausal.

The Analogy: Imagine a classroom of 100 students. If you ask them, "Who has finished high school?" 70 raise their hands. But if you check the principal's official attendance log, it only says 10 students have graduated.
In this study, the survey found seven times more menopausal women than the medical records did. The medical records were essentially "missing" the story for most women.

Why Was the Medical Record So Quiet?

The researchers dug deeper to see why the doctors didn't write it down. They found it wasn't because the women had different types of menopause (natural vs. surgical). It was simply that doctors often forget to check the box.

• The "Billing" Problem: Doctors usually only write down a diagnosis if they need it to justify a specific treatment or insurance claim. If a woman comes in for a sore throat and mentions her periods stopped years ago, the doctor might not type "menopause" into the computer because it's not relevant to the sore throat.
• The Missing Symptoms: The study also found that common menopause symptoms (like hot flashes or mood swings) were almost invisible in the medical records. It's like a weather report that only records "rain" but never mentions "humidity" or "wind," even though those are what people actually feel.

The "Age 65" Mystery

When looking at the ages of women in the medical records, the data looked a bit weird. There was a huge spike in menopause diagnoses right at age 65.

The Analogy: Think of a highway. Usually, traffic flows smoothly. But at age 65, there's a massive traffic jam of "menopause diagnoses."
The researchers suspect this is because Medicare (health insurance for seniors) starts at 65. When women turn 65, they get a new insurance card, and doctors might be more likely to update their records or run more check-ups, causing a sudden flood of data. It's likely an artifact of the insurance system rather than a biological event where everyone hits menopause at exactly 65.

The "Three-Way" Intersection: A Rare Gem

The study also looked at how many women had data in all three categories:

1. Medical Records
2. Surveys
3. Genetic Data (DNA samples)

While the survey had lots of women, and the medical records had some, the group of women who had all three was much smaller (about 22,000).
The Analogy: Imagine trying to find a specific type of rare coin. You have a bag of 193,000 coins (Surveys), a bag of 28,000 coins (Medical Records), and a bag of 250,000 coins (DNA). But when you try to find the coins that are in all three bags at the same time, you only have a small handful left. This means future studies that want to combine DNA with medical history will have to work with smaller groups of people.

The "Response Bias" (The 70-Year-Olds Who Said "No")

The researchers noticed something odd in the survey data. About 4% of women over the age of 70 said, "No, my periods haven't stopped."
The Reality Check: Biologically, it is extremely rare for a woman to still have periods at 70. This suggests that some older women might have just clicked "No" by mistake, forgot, or didn't want to answer the question. It's like a census where some people claim they are still in kindergarten at age 60.

What Does This Mean for the Future?

This paper is a "User Manual" for scientists who want to study menopause using the All of Us data.

1. Don't rely only on medical records: If you want to study menopause, you must use the surveys. The medical records are too quiet and will miss most of the story.
2. Be careful with age: Remember that the "Age 65 spike" in medical records is likely due to insurance changes, not biology.
3. Plan your sample size: If you want to mix DNA, medical records, and surveys, you need to know that your group of participants will be much smaller than if you just used the surveys.

The Bottom Line:
The All of Us program is a treasure trove of data, but the "Menopause" section of the medical record is currently under-stocked. By understanding these gaps, scientists can design better studies to understand how menopause affects heart health, bones, and genetics, ultimately leading to better healthcare for women of all backgrounds.

Technical Summary

1. Problem Statement

Menopause is a critical physiological transition associated with significant health risks, particularly cardiovascular, metabolic, and osteoporotic diseases. However, large-scale precision medicine research often suffers from:

• Inconsistent Data Capture: Menopause status is frequently missing, incomplete, or inconsistently measured in Electronic Health Records (EHRs).
• Lack of Granularity: There is a gap in understanding how menopause timing (age) and type (natural vs. surgical) vary across diverse sociodemographic groups.
• Data Silos: Few studies integrate EHR diagnostic codes, self-reported survey data, and genomic data to characterize the menopause transition in a diverse US population.

The authors aim to characterize menopause-related data within the All of Us Research Program (AoURP) to inform future study designs, power calculations, and phenotype definitions.

2. Methodology

The study utilized the All of Us Research Program Controlled Tier datasets (v8 and v7).

• Cohort: Approximately 396,000 participants identified as female (based on sex at birth).
• Data Modalities Analyzed:
  1. EHR Data: Structured diagnostic codes using SNOMED CT vocabulary. Key codes included:
     • Menopause present (289903006)
     • Premature menopause (373717006)
     • Ovarian failure types (e.g., 237138004, 65846009)
     • Abnormal vasomotor function (70670009)
  2. Survey Data: Responses from the "Women's Health" section of the Overall Health survey, specifically:
     • "Have your menstrual periods stopped permanently?"
     • Reason for cessation (natural, surgery, medication, etc.).
     • History of hysterectomy and oophorectomy.
  3. Genomic Data: Short-read whole-genome sequencing (srWGS) data.
• Sociodemographic Variables: Race, ethnicity, education, income, and genetic ancestry.
• Age Grouping: Participants were stratified into:
  • Pre-menopausal (<40 years)
  • Peri-menopausal (40–60 years)
  • Post-menopausal (>60 years)
• Privacy Protocols: Sample sizes ≤20 were obscured; larger counts were rounded to the nearest 5 to prevent re-identification.
• Analysis: Descriptive statistics (counts, medians, IQRs) and intersection analysis (Venn diagrams logic) were performed using SQL, R, and Python to assess overlap between modalities.

3. Key Contributions

• First Comprehensive Characterization: This is the first large-scale descriptive summary of menopause data specifically within the AoURP cohort across EHR, survey, and genomic modalities.
• Quantification of Under-ascertainment: The study provides concrete evidence that EHRs significantly under-detect menopause compared to self-reporting.
• Sociodemographic Stratification: It evaluates how menopause data availability and age distributions vary across race, ethnicity, and genetic ancestry groups.
• Guidance for Future Research: The paper offers specific recommendations for researchers on how to define phenotypes, calculate power, and handle data limitations when studying menopause in AoURP.

4. Key Results

A. Discrepancy Between EHR and Survey Data

• Volume Difference: Survey data captured menopause observations in ~193,000 participants, whereas EHR diagnostic codes captured only ~28,000. This represents a ~7-fold difference, indicating severe under-ascertainment in structured EHR data.
• Concordance: Nearly 99% of participants with an EHR menopause diagnosis also reported menopause in the survey. However, the vast majority of survey-reported menopause cases were not captured in the EHR.
• Symptom Gap: Common menopause symptoms, such as vasomotor symptoms (hot flashes), were almost entirely absent from structured EHR data (<20 observations), highlighting a major documentation gap.

B. Age Distributions and Patterns

• Survey Data: Followed biological expectations.
  • <40 years: Predominantly pre-menopausal.

  • 60 years: Predominantly post-menopausal.

  • Bias Note: A small subset (~4% of those >70 years) reported no menopause, suggesting potential recall or response bias.
• EHR Data: Showed a flatter distribution with a notable spike at age 65. This suggests a reporting artifact, potentially linked to Medicare eligibility or insurance billing practices rather than biological onset.

C. Intersection of Modalities

• Genomic Overlap: Among ~250,000 females with genomic data:
  • ~99% had survey menopause data.
  • Only ~9% had EHR menopause data.
  • Consequently, only ~9% of the genomic cohort had intersecting EHR and survey menopause data, limiting the power for three-way integrated analyses.

D. Sociodemographic Variations

• Minor shifts in age distributions were observed across race and ancestry groups (e.g., White/European-ancestry participants were slightly older on average).
• The authors conclude these differences largely reflect the underlying age distribution of the AoURP cohort rather than specific biological differences in menopause timing across groups.

5. Significance and Implications

• Study Design: Researchers utilizing AoURP for menopause studies should prioritize survey data over EHR codes for phenotype definition due to the 7x higher capture rate.
• Clinical Documentation: The findings highlight a critical need for improved systematic documentation of menopause status and symptoms in EHRs to support clinical care and research.
• Precision Medicine: The study establishes a foundation for future genomic studies. While the three-way intersection (EHR + Survey + Genomics) is currently small, the large survey-based cohort allows for robust genetic association studies regarding menopause timing and symptoms.
• Health Equity: By characterizing data availability across diverse groups, the study ensures that future menopause research can be designed to be representative of the US population, addressing historical gaps in women's health research.

Conclusion: The paper concludes that while AoURP offers a unique, diverse resource for menopause research, reliance on EHR codes alone will lead to significant selection bias and underpowered studies. A multimodal approach, heavily weighting self-reported survey data, is essential for accurate characterization of the menopause transition.