📄 obstetrics and gynecology

Modeling cycle phases using hormone trajectories in women with and without polyendocrine metabolic ovarian syndrome

This study utilizes self-tracked urinary hormone data to develop an autoregressive Hidden Markov model that accurately maps menstrual cycle phases, revealing distinct hormonal differences in women with polyendocrine metabolic ovarian syndrome (PMOS) and demonstrating the model's potential for non-invasive PMOS screening and improved clinical testing protocols.

Original authors: Stujenske, T. M., Bouchard, T. P., Troy, A., Kelemen, S., Folino, B., Wills, T., Sugden, L. A.

Published 2026-06-04

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Stujenske, T. M., Bouchard, T. P., Troy, A., Kelemen, S., Folino, B., Wills, T., Sugden, L. A.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine your menstrual cycle as a complex, 28-day road trip. For decades, doctors have tried to map this journey by checking your location based on the calendar date—like saying, "On day 14, you should be at the Grand Canyon." But the problem is, everyone's trip is different. Some people drive fast, some drive slow, and some get stuck in traffic. If you only check the calendar, you might miss the Grand Canyon entirely because you arrived on day 12 or day 18.

This paper introduces a smarter way to map that trip using a "GPS" that doesn't care about the date, but instead cares about the scenery (your hormone levels) you are seeing right now.

Here is the breakdown of their study in simple terms:

1. The Problem: The Calendar is a Bad Map

The researchers looked at data from over 1,100 women who used a home device (the Mira monitor) to test their urine for three key hormones:

E13G: A form of estrogen (the "building" hormone).
LH: Luteinizing hormone (the "trigger" that signals ovulation).
PDG: A form of progesterone (the "maintenance" hormone after ovulation).

The old way of doing things is to tell women, "Test your hormones on day 13, 14, and 15." But because cycles vary wildly in length and timing, this is like telling a driver to look for a specific mountain at 2:00 PM. If they are driving a different route, they might miss it.

2. The Solution: The "Smart GPS" (arHMM)

The team built a computer model called an autoregressive Hidden Markov Model (arHMM). Think of this as a smart GPS that ignores the clock and only looks at the road signs.

How it works: Instead of asking "What day is it?", the model asks, "What is the hormone doing right now?" Is it rising? Is it falling? Is it staying flat?
The Six Zones: The model divides the cycle into six distinct "zones" or phases based on these hormone movements, rather than calendar days:
1. Early Follicular: The reset button (hormones are low).
2. Late Follicular: The ramp-up (estrogen starts climbing).
3. Periovulatory: The peak (LH surges like a rocket).
4. Early Luteal: The immediate aftermath (estrogen drops, progesterone starts rising).
5. Mid-Luteal: The plateau (progesterone is high).
6. Late Luteal: The wind-down (hormones drop before the next period).

Because the model follows the shape of the hormone curve, it works perfectly for women with short cycles, long cycles, or irregular cycles.

3. What They Found: The "PMOS" Difference

The researchers compared women with healthy cycles to women who self-reported having PMOS (Polyendocrine Metabolic Ovarian Syndrome, formerly known as PCOS). They found distinct differences in how the "GPS" mapped their trips:

The Missing Surge: In healthy women, the "LH Trigger" (Phase 3) is a massive, sharp spike. In women with PMOS, this spike was much weaker and lower.
The Weak Maintenance: After the trigger, healthy women have a strong, steady rise in progesterone (Phase 5). Women with PMOS had much lower levels here, suggesting their "engine" wasn't running as smoothly after the trigger.
The Long Wait: Women with PMOS spent significantly more time in the "ramp-up" phase (Phase 2) before the trigger ever happened. It's like they were stuck in traffic for days before finally getting to the highway.

4. Why This Matters (According to the Paper)

The study tested two ways of checking hormones:

The Calendar Way: Checking on fixed days (e.g., days 13–16).
The GPS Way: Checking when the model says you are in the specific "LH Surge" phase.

The Result: The GPS way was much better at catching the actual hormone events. The calendar way often missed the peak entirely because the woman's cycle was just a few days off schedule.

5. The "Detective" Test

Finally, the researchers used the data from this GPS model to see if they could tell if a woman had PMOS just by looking at her hormone patterns.

They fed the model's findings (like "weak LH spike" or "long ramp-up time") into a computer classifier.
The Score: The computer could correctly identify whether a woman had PMOS or was healthy about 78% of the time.

The Bottom Line

This paper doesn't claim to cure PMOS or replace a doctor's diagnosis. Instead, it shows that by using a "smart GPS" to track hormone movements instead of just counting calendar days, we can:

Get a much clearer picture of what is actually happening inside the body.
Spot the specific "traffic jams" and "weak engines" that happen in PMOS.
Potentially screen for these conditions using non-invasive home testing, rather than waiting for a doctor to guess the right day to run a blood test.

It turns the menstrual cycle from a confusing calendar puzzle into a readable map of your body's unique rhythm.

Technical Summary: Modeling Cycle Phases Using Hormone Trajectories in Women with and without Polyendocrine Metabolic Ovarian Syndrome (PMOS)

Problem Statement
Menstrual-related disorders, such as polyendocrine metabolic ovarian syndrome (PMOS, formerly polycystic ovary syndrome), affect approximately 10% of reproductive-age women. A significant barrier to early detection and characterization of these disorders is the reliance on fixed-cycle-day clinical protocols (e.g., testing on days 13–16 or 19–22). These protocols fail to account for the substantial inter-individual variability in cycle length and the timing of physiological events, particularly in women with irregular cycles. Furthermore, traditional models based on average cycle parameters are inaccurate for predicting phases in irregular cycles. While at-home urinary hormone tracking technologies (e.g., the Mira Monitor) now provide high-resolution quantitative data, there is a lack of computational frameworks capable of modeling day-to-day hormonal dynamics to define physiologically meaningful phases across diverse populations.

Methodology
This study utilized a secondary analysis of self-tracked urinary reproductive hormone data (Estrone-3-glucuronide [E13G], Luteinizing Hormone [LH], and Pregnanediol Glucuronide [PDG]) collected from over 1,100 individuals (277 healthy controls and 841 with self-reported PMOS) via the Mira Monitor. The dataset comprised over 3,800 menstrual cycles.

The core methodological innovation was the development of an autoregressive Hidden Markov Model (arHMM) to map cycle days to latent physiological phases. The approach involved:

Data Curation and Indexing: A curated set of healthy control cycles (24–38 days) was used to train the model. Three indexing strategies were employed to align data: forward indexing (Day 1 = bleeding start), backward indexing (Day -1 = cycle end), and LH-indexing (Day 0 = LH surge peak).
Phase Definition: Six distinct phases were defined based on the directional trajectory of hormones (rising, falling, or constant):
- Phase 1: Early follicular (Days 1–5).
- Phase 2: Late follicular (10–2 days prior to LH surge).
- Phase 3: Periovulatory (2–0 days prior to LH surge).
- Phase 4: Early luteal (0–4 days post-LH surge).
- Phase 5: Mid-luteal (9–5 days before cycle end).
- Phase 6: Late luteal (Last 5 days).
Model Architecture: The arHMM utilized a transition matrix enforcing sequential progression through phases 1–6. The emission model was autoregressive, predicting the hormone level on day $i$ ( $X_i$ ) based on the level on day $i-1$ ( $X_{i-1}$ ) and the current latent phase ( $Y_i$ ). This allowed the model to capture day-to-day trajectories (slopes) rather than absolute hormone levels, accommodating variability in absolute concentrations between individuals.
Feature Extraction and Classification: Individual-level features were extracted from the model output, including hormone statistics (min, max, mean, median, SD) within each phase, phase durations, and model fit scores. A Random Forest classifier was trained on these features to distinguish between individuals with PMOS and healthy controls.
Ovulation Proxy: Cycles were labeled ovulatory or anovulatory based on a threshold analysis of PDG levels in Phase 5 relative to Phase 3 baselines.

Key Results

Model Performance: The arHMM successfully reproduced expected hormonal trajectories across the six phases, including E13G rises in late follicular/periovulatory phases, LH surges in the periovulatory phase, and PDG rises in the luteal phase.
Phase-Based vs. Cycle-Day Testing: Comparing hormone levels in arHMM-derived phases versus fixed calendar days revealed that phase-based testing significantly outperformed fixed-day protocols. Specifically, maximum LH levels in Phase 3 and mean PDG levels in Phase 5 were significantly higher than those captured on standard clinical days (13–16 and 19–22, respectively; $p < 0.001$ ), suggesting fixed-day protocols often miss peak physiological events.
PMOS Phenotyping: Distinct phase-specific differences were identified between healthy controls and the PMOS group:
- LH: Lower LH levels in Phase 3 (periovulatory) for the PMOS group.
- PDG: Reduced PDG levels in Phases 5 and 6 (mid- and late-luteal) for the PMOS group.
- E13G: Generally higher E13G levels across phases in the PMOS group, though patterns were heterogeneous.
- Phase Duration: PMOS cycles spent significantly more time in Phase 2 (late follicular) compared to controls ( $p < 0.001$ ), consistent with oligomenorrhea.
Classification Accuracy: Using arHMM-derived hormone and phase length features, the Random Forest classifier achieved 78.4% accuracy in distinguishing PMOS status. Including cycle length features slightly improved accuracy to 79.2%.
Ovulatory Status: Cycles in the PMOS group were less likely to meet the PDG-based thresholds for ovulation compared to healthy controls, aligning with clinical expectations of anovulatory dysfunction in PMOS.

Significance and Claims
The paper claims that the arHMM framework offers a novel, biologically grounded approach for modeling menstrual cycle dynamics that overcomes the limitations of fixed-day clinical protocols. By focusing on directional trajectories rather than absolute levels, the model accommodates the natural variability in cycle length and ovulation timing.

The authors assert that this approach demonstrates the potential for:

Non-invasive Screening: Providing a method to screen for PMOS and characterize reproductive health using self-tracked data with ~78% accuracy.
Improved Clinical Protocols: Highlighting the limitations of fixed-day testing and suggesting that phase-based timing could yield more accurate hormonal assessments, particularly for individuals with irregular cycles.
Physiological Insight: Enabling the identification of latent sub-phases and distinct hormonal signatures associated with PMOS, such as reduced luteal progesterone and periovulatory LH deficits.

The study concludes that integrating probabilistic modeling with home-based quantitative hormone tracking represents a step toward scalable, personalized reproductive health monitoring, though it acknowledges the need for future validation in larger, diverse cohorts with ultrasound-confirmed ovulation and well-defined PMOS phenotypes.