I entered perimenopause at 38. For the next 13 years, I lived in a body that felt like it was constantly betraying me—two-week long periods, relentless fatigue, and a level of bleeding that made ordinary life feel like a logistical puzzle with no solution. I sweat through my suits on stage and in boardrooms, drenched the sheets at night, and bled through clothing often enough to plan my days around proximity to a bathroom and a backup outfit. One physician suggested uterine ablation as if the only goal was to “solve” the mess. Another told me it was all in my head. Neither offered what I needed most: a framework that took my symptoms seriously and a care plan grounded in the lived reality of midlife women.
If you’ve ever felt dismissed in perimenopause, you’re not alone—and the reason is bigger than any individual clinician. We have a data problem.
Perimenopause is the menopausal transition—often years long—when hormonal fluctuations can produce hot flashes, night sweats, irregular or heavy bleeding, mood shifts, sleep disruption, and more. The Menopause Society notes that vasomotor symptoms (hot flashes and night sweats) can affect up to 80% of women during the transition. (The Menopause Society) And yet, midlife women routinely discover that the healthcare system has gaps in training, guidelines, and evidence—especially for those of us who don’t match the “average” patient profile on which so much historical research was built.
As an AI ethicist, I think about this constantly: if women’s midlife health has been under-measured, mischaracterized, or generalized from the wrong populations, what happens when we train models on that legacy data? We don’t get neutral technology. We get scalable neglect.
Harm is what happens when “the data” doesn’t include you—or mislabels you
A vivid example of how women can be harmed by misapplied evidence is the hormone therapy whiplash that followed the Women’s Health Initiative (WHI). In July 2002, the NIH halted the combined estrogen-plus-progestin arm early because the balance of risks and benefits in that study population did not support continuation. (Fred Hutch) That moment mattered. But what happened next is the ethical lesson: a specific dataset, drawn from a specific cohort, was treated as universally predictive for all women.
The WHI hormone trials enrolled postmenopausal women ages 50–79. (whi.org) When results landed in the public imagination, nuance collapsed. Many symptomatic women—often younger, closer to the onset of menopause, and seeking relief rather than “chronic disease prevention”—were swept into a single risk narrative that wasn’t designed for them.
You can see the downstream impact in prescribing behavior. A JAMA analysis of prescribing after WHI reported that standard-dose Prempro use fell 64%, from 4.4 million prescriptions in Q2 2002 to 1.6 million in Q1 2003, and overall hormone therapy prescriptions declined 32% within nine months of the WHI estrogen+progestin report. (JAMA Network) That isn’t an abstract policy debate; it’s millions of women experiencing symptoms while clinicians and patients navigated fear—sometimes with incomplete context.
Fast forward to November 10, 2025: the FDA announced it was initiating removal of broad boxed warnings from menopausal hormone replacement therapy (HRT) products, following review, an expert panel, and public comment. (U.S. Food and Drug Administration) Whether you agree with every characterization in that announcement, it underscores a central truth: labels, warnings, and guidelines can become “frozen” artifacts of old data and old interpretations—long after the science and clinical practice evolve. (U.S. Food and Drug Administration)
And HRT is not the only place where “one-size thresholds” have missed women.
Consider cardiology. A landmark prospective cohort study (published in BMJ) evaluated suspected acute coronary syndrome patients (n=1126, 46% women) and compared a contemporary single troponin threshold (50 ng/L) with a high-sensitivity troponin I assay using sex-specific thresholds (34 ng/L for men; 16 ng/L for women). (Australian Women’s Health Alliance) The results were striking: the diagnosis of myocardial infarction in women rose from 11% to 22% when sex-specific thresholds were applied, effectively doubling identification in women, while changes in men were comparatively small (19% to 21%). (Australian Women’s Health Alliance) This is what “data disparity” looks like in practice: women aren’t having fewer heart attacks; they’re being recognized later—or not at all—because the measurement standard wasn’t calibrated for them.
Pharmacology has similar scars. In 2013, the FDA approved new label changes and sex-specific dosing recommendations for zolpidem (Ambien products) after recognizing next-morning impairment risk. The FDA stated the recommended initial dose for certain immediate-release products is 5 mg for women (vs 5 or 10 mg for men), and for extended-release (Ambien CR) 6.25 mg for women (vs 6.25 or 12.5 mg for men). (U.S. Food and Drug Administration) When dosing guidance is built from male-skewed pharmacokinetic assumptions, women pay the price in adverse events, impaired functioning, and safety risks.
Even outside healthcare, gendered data gaps harm women. In late 2025, the U.S. Department of Transportation highlighted the THOR-05F—an advanced female crash test dummy—aimed at addressing longstanding disparities in vehicle safety testing. (Department of Transportation) Coverage of the announcement cited estimates that women are 73% more likely to be injured and 17% more likely to die in crashes compared with men in comparable collisions. (AP News) When the “standard body” in testing is male, women’s bodies become the edge case.
The training gap is real—and it shows up in midlife care
One reason perimenopause is so often mishandled is that clinicians aren’t consistently trained for it.
A survey of U.S. residency program directors (OB-GYN, family medicine, internal medicine) reported that only 31.3% said their program had any menopause curriculum. Of those, 71% reported two or fewer menopause lectures, and 83.8% agreed they needed more menopause training resources. (The Menopause Society) Earlier work found similarly limited formal curricula: one cited figure was 20.8% of programs reporting a formal menopause curriculum in 2013. (The Menopause Society)
Resident knowledge assessments also reveal uneven preparation. In one multi-specialty survey effort (703 residents contacted; 183 responses; 26% response rate), correct answers on menopause questions clustered around 53% in OB-GYN, 56% in family medicine, and 23% in internal medicine. Those numbers help explain why so many women—especially those of us with complex bleeding patterns, severe vasomotor symptoms, or mood and sleep disruption—feel like we’re teaching our providers in real time.
Now layer in the question you asked me directly: what percentage of clinicians are board certified in menopause? The Menopause Society has expanded its clinician certification footprint, but the scale still illustrates the gap. The AAMC reported that the number of Menopause Society Certified Practitioners (MSCPs) rose from about 1,350 in 2021 to about 4,100 in 2025. (AAMC) Put that beside the size of the U.S. physician workforce—HRSA reports 839,108 active patient care physicians (U.S.). (Bureau of Health Workforce) Even if every MSCP were a physician (they’re not—many are NPs and other clinicians), 4,100 is still under 0.5% of U.S. patient care physicians. (AAMC) That’s not a criticism of the certifying body; it’s a measurement of unmet need.
Why AI models built on old data will replicate these harms—at scale
If medicine has historically treated midlife women as an afterthought, AI will not magically fix that. It will encode it.
In machine learning, we talk about dataset shift and model drift: when the data distribution or clinical context changes over time, a model trained on yesterday’s world can quietly degrade—sometimes catastrophically. A 2025 JAMA Network Open study on clinical AI deployment described proactive monitoring approaches to detect and mitigate harmful data shifts, emphasizing that shifts can undermine model performance and patient safety. (JAMA Network) Reviews in clinical ML likewise warn that dataset shift is common in healthcare and requires active detection and governance. (OUP Academic)
Now apply that to women’s midlife health:
- If the “ground truth” labels in historic records underdiagnose women’s cardiac events because thresholds were not sex-specific, then models trained on those labels learn that bias as if it were biology. (Australian Women’s Health Alliance)
- If prescribing patterns cratered after WHI and stayed suppressed for years, observational data may reflect access barriers and fear—so a model might infer that fewer women “need” therapy, rather than recognizing fewer women receive it. (JAMA Network)
- If menopause education is inconsistently taught, then clinical notes will vary wildly by specialty and training exposure—creating noisy features and missingness that models misinterpret as patient variation. (The Menopause Society)
This is not hypothetical. Fairness and bias mitigation have become central themes in clinical ML ethics precisely because models can magnify existing inequities when trained on biased data and deployed without rigorous monitoring. (Nature)
As a woman who spent more than a decade bleeding, sweating, and suffering while being minimized, I don’t want an “AI for women’s health” that simply automates the same dismissal I experienced—only faster, cheaper, and harder to challenge.
The (M) Factor 2.0: Before The Pause—and why this moment matters
I’m part of The (M) Factor 2.0: Before The Pause (Perimenopause)—a film that shines light on the most disruptive and misunderstood phase of hormonal change and argues that early awareness can transform health and identity for millions of women. (creativevisions.org) The reason storytelling matters in a data conversation is simple: what isn’t measured rarely gets modeled, funded, taught, or prioritized. Perimenopause has been treated as background noise for too long, and that silence has consequences—clinical, economic, and personal.
If you’re searching for The (M) Factor 2.0: Before The Pause, perimenopause support, AI for women’s health, or Susan Sly, I want you to find this message: we deserve modern medicine built from modern, representative data—and modern AI governed accordingly.
Because here is what I know, both from lived experience and from ethical analysis: when the system insists women’s midlife suffering is “just stress,” “normal aging,” or “in your head,” it is not only failing compassion. It is failing epistemology—failing at knowing.
A better standard for women’s midlife health data—and for women’s health AI
We cannot build safe models with old, incomplete, or misapplied data. Full stop. And “old” doesn’t just mean time; it means misaligned cohorts, outdated thresholds, and biased labels.
For women’s midlife health, the ethical path forward is to treat data as a living clinical asset: continuously updated, disaggregated by sex and relevant subgroups, transparently documented, and actively monitored after deployment. (Grants.gov) This aligns with NIH’s expectation that sex as a biological variable be factored into research design, analysis, and reporting—because rigor and relevance depend on it. (Grants.gov)
And it means something practical for patients: if your symptoms are severe, if your bleeding is heavy or prolonged, if you’re soaking sheets or missing work or feeling like a stranger in your own skin—you are not “too sensitive.” You are a signal in a system that has historically treated women’s data as optional. We are changing that.
I spent 13 years paying the price of a gap between women’s bodies and women’s evidence. In the era of AI, we don’t get to make that mistake again—because the next time, the harm won’t be one doctor in one exam room. It will be an algorithm everywhere.
Medical note: This essay is educational and not medical advice. If you have heavy bleeding, new bleeding patterns, or concerning symptoms, please seek care from a qualified clinician.
About Susan Sly
Susan Sly is the maven behind Raw and Real Entrepreneurship. An award-winning AI entrepreneur and MIT Sloan alumna, Susan has carved out a niche at the forefront of the AI revolution, earning accolades as a top AI innovator in 2023 and a key figure in real-time AI advancements for 2024. With a storied career that blends rigorous academic insight with astute market strategies, Susan has emerged as a formidable founder, a discerning angel investor, a sought-after speaker, and a venerated voice in the business world. Her insights have graced platforms from CNN to CNBC and been quoted in leading publications like Forbes and MarketWatch. At the helm of the Raw and Real Entrepreneurship podcast, Susan delivers unvarnished wisdom and strategies, empowering aspiring entrepreneurs and seasoned business veterans alike to navigate the challenges of the entrepreneurial landscape with confidence.
Connect With Susan
- Website: https://thepause.ai/
- Website: https://susansly.com
- LinkedIn:@susansly
