“I just don’t feel like myself lately.”
It is one of the most common opening lines in midlife primary care, and one of the least diagnostically useful on its face. The patient is forty-four, sleeping badly, irritable in a way she describes as unfamiliar, and struggling to hold a thought through a meeting. Her cycles are still coming, roughly. She has read about perimenopause and dismissed the idea, because she is not having hot flashes.
Nearly four in ten women say the encounter that follows gets it wrong.
The size of the gap
A 2025 Biote Corporation survey of 1,005 women aged 30 to 60 found that 39 percent felt they had been misdiagnosed when seeking care for perimenopausal symptoms. Anxiety and depression were commonly treated as free-standing conditions rather than assessed alongside the hormonal transition that may have been driving them.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The clinician side of the picture is equally instructive. Only 42 percent of clinicians initiated a discussion about perimenopause, and just 15 percent of women felt adequately informed when their symptoms began. Among women aged 30 to 44, 46 percent were never proactively asked whether they had experienced perimenopausal symptoms, and only 26 percent received information from their primary care physician or gynecologist.
What women actually report
The mismatch at the center of this is specific and measurable. In a 2026 peer-reviewed global digital survey of 17,494 women across 158 countries published in Menopause, the following pattern emerged among the 12,681 respondents aged 35 and older:
Fatigue and mental exhaustion affect 83 percent of women in the transition. Severe irritability affects 80 percent. Depressive mood and anxiety affect 77 percent. Meanwhile 71 percent anticipate that hot flashes will be the hallmark that tells them perimenopause has begun.
Both the patient and the clinician are, in effect, waiting for a symptom that arrives late or not at all, while the symptoms that do arrive map neatly onto a psychiatric differential. The prescription that follows is often reasonable on the information available. The information available was incomplete.
Why FSH testing rarely settles it
Clinicians reaching for a laboratory answer usually reach for follicle-stimulating hormone, and in the perimenopausal window it performs poorly. The test gets ordered because it is what we have, the result comes back normal, and the patient leaves the office more confused than when she arrived. FSH fluctuates substantially cycle to cycle and within a single cycle during the transition, which means a value drawn on an arbitrary day describes that day rather than the patient’s stage. A normal result in a symptomatic forty-six-year-old is common and reassures nobody correctly.
The Stages of Reproductive Aging Workshop +10 (STRAW+10) criteria remain the accepted staging framework, and they anchor the diagnosis in menstrual cycle changes rather than in hormone values. FSH is a supporting datum for late reproductive and late menopausal transition stages, not a decisive one, and STRAW+10 explicitly cautions against relying on it in the early transition where variability is highest.
The diagnosis in a woman over forty-five with characteristic symptoms and menstrual change is clinical. Laboratory testing earns its place in a narrower set of circumstances: suspected primary ovarian insufficiency under forty, an unclear picture where thyroid disease or anemia needs exclusion, or a patient on hormonal contraception where cycle history is uninformative.
Working through the differential
| Presentation | Points toward perimenopause | Points elsewhere |
|---|---|---|
| Low mood and anxiety | New in the forties, fluctuating rather than persistent, tracks with cycle phase, accompanied by sleep disruption and cycle change | Present since the twenties, persistent rather than cyclical, anhedonia prominent, clear psychosocial precipitant |
| Fatigue | Worse after disrupted nights, improves on good sleep, coincides with cycle irregularity | Progressive regardless of sleep, weight change, cold intolerance, heavy menstrual bleeding |
| Cognitive complaints | Word-finding and working memory, fluctuating, no functional decline | Progressive, functional impairment, family concern, disorientation |
| Palpitations | Episodic, often nocturnal, no exertional pattern | Exertional, syncope, family history of sudden death |
| Joint pain | Diffuse, morning stiffness under 30 minutes, new in the transition | Synovitis, prolonged stiffness, inflammatory markers raised |
| Non-restorative sleep and daytime somnolence | Fragmented sleep with 2 to 4 a.m. wakening, tracks with vasomotor or hormonal symptoms, partner reports quiet sleep | Loud snoring, witnessed apneas, morning headache, resistant hypertension, elevated neck circumference, high STOP-BANG; consider obstructive sleep apnea |
Thyroid function, ferritin, and a full blood count belong in almost every one of these workups. Their purpose is not to diagnose perimenopause but to remove the conditions that imitate it, and the two questions are frequently conflated.
How AI solves a data integration problem
The problem is not that we lack the knowledge to diagnose perimenopause well. The problem is that the knowledge is scattered across places our current systems do not connect: cycle-tracking apps the patient uses, wearable sleep and heart-rate data she checks daily, lab values from three different practices over four years, and a memory of how she felt eighteen months ago that no chart captures. The mismatch between the symptoms women experience and the diagnoses they receive is, at its root, a data integration problem dressed up as a clinical one.
A clinical decision support layer designed for this stage does five things that a 15-minute appointment structurally cannot.
- It assembles the trajectory before the visit begins. Perimenopause is defined by variability, which makes it structurally difficult to assess in the format we usually assess things: one appointment, one snapshot, one set of values. Cycle history, symptom logs, wearable-derived sleep architecture, resting heart rate drift, prior thyroid and ferritin and glucose values, and prescribing history are pulled into a single longitudinal view. What appears in the clinician’s summary is not the most recent number but the direction the numbers have been moving. A patient whose sleep has degraded over eighteen months, whose resting heart rate has drifted upward, and whose symptom pattern has become cyclical is describing a trajectory. The trajectory is the finding.
- It stages the patient against a recognized framework. STRAW+10, for example, gives us a reproducible way to describe where a woman is in the transition based on cycle characteristics, symptom pattern, and, where relevant, biomarker context. The infrastructure automates the staging so the clinician can confirm or override it rather than reconstruct it from scratch every visit. The physician remains the decision-maker; the platform removes the busywork.
- It surfaces the correlations that a single visit hides. When mood symptoms cluster in the luteal phase, when sleep fragmentation predicts the next irritable day, when a new joint complaint tracks with an emerging vasomotor pattern, these relationships live in the data. The infrastructure connects them, cites the peer-reviewed evidence behind the connection, and hands the pattern to the physician as a starting point for the conversation, not a conclusion.
- It flags the differentials that need to be ruled out. Obstructive sleep apnea, thyroid disease, iron deficiency, primary ovarian insufficiency, and, where relevant, the mood disorders that genuinely coexist with the transition rather than replace it as the diagnosis. Each flag traces back to the specific data point or symptom pattern that triggered it, and each is proposed to the physician as a hypothesis rather than declared as a finding. Physician-in-the-loop is not a caveat; it is the design.
- It gives the patient her own picture back. The same longitudinal record the clinician sees at the start of the visit is available to the patient in language she can read. What has changed. What has not. What we are watching. What we are ruling out. What the next check-in will look at. A woman who has spent three years being told her labs are normal experiences a different kind of visit when the record can show her what her labs have been doing, cycle by cycle, year over year. Feeling heard is downstream of feeling seen. Being seen requires infrastructure that can hold what a woman is experiencing across the years, not the 15 minutes.
None of this replaces the clinician. All of it changes what the 15-minute visit can accomplish. The attending decides. The infrastructure ensures the attending walks in with the whole picture rather than the most recent slice of it. In a diagnosis defined by direction of travel rather than by threshold values, that difference is the difference between the visit that gets it wrong and the visit that gets it right.
The solution
This is a large part of why we built the AI clinical decision support tool Longevitix around continuity rather than around the visit. Symptom reports, laboratory results across years, prescribing history, and wearable-derived sleep and heart-rate trends are held as a single longitudinal record, and the summary a clinician sees at the start of an appointment reflects the direction of travel rather than the most recent value. Every conclusion the platform surfaces traces back to the data and the cited evidence behind it, which matters most in a diagnosis that is made from pattern rather than from a threshold.
For the patient who opens with “I don’t feel like myself lately,” a record that can show what she looked like eighteen months ago is worth more than any single test.
This essay is cited in the KevinMD record on medical errors.
Neil Panchal is a board-certified emergency physician and the cofounder and chief medical officer of Longevitix, where he builds the clinical and AI infrastructure behind precision longevity medicine.
He trained in emergency medicine at Mount Sinai in New York City and completed agentic AI training with the Harvard Data Science Institute and Stanford Medicine. His clinical affiliations include Yale New Haven Health, Hackensack Meridian Health, and Atlantic Health. He has spent his career where acute medicine meets prevention, treating the downstream failures of chronic disease that earlier risk stratification could have prevented.
At Longevitix, he directs the clinical evidence engine. He translates clinical research, biomarkers, wearables, and multimodal data into personalized protocols, and he sets the standards that govern how AI applies to physician workflows and patient care. His work spans healthspan optimization, early disease detection, and clinical decision support, pairing frontline medical judgment with health informatics.
He writes and speaks on evidence-based longevity, preventive medicine, and the responsible deployment of AI in clinical practice, with recent work including peer-reviewed research, commentary on physician oversight of agentic AI, and writing on KevinMD. He also publishes on the Longevitix blog and shares updates on LinkedIn, where Longevitix maintains a company page as well.
