Computational Protocols for Menstrual Cycle Phase Classification: A Guide for Biomedical Research and Drug Development

Abstract

This article provides a comprehensive guide for researchers and drug development professionals on computational protocols for determining menstrual cycle phases. It covers the foundational rationale for moving beyond error-prone self-report methods, details the application of machine learning with multi-modal physiological data from wearables, addresses troubleshooting for irregular cycles and data variability, and establishes frameworks for rigorous model validation. By synthesizing current research, this resource aims to enhance methodological rigor in studies of female physiology, leading to more reliable data for clinical trials and women's health innovation.

The Critical Foundation: Why Accurate Menstrual Cycle Phase Determination Matters in Research

{ARTICLE CONTENT BEGINS HERE}

Limitations of Traditional Counting and Self-Report Methods

Within reproductive biology and drug development research, the precise determination of menstrual cycle phase is critical for investigating cycle-dependent physiological changes, pharmacokinetics, and therapeutic outcomes. For decades, researchers have heavily relied on traditional methods, primarily self-reported cycle length and calendar-based counting, to assign participants to specific cycle phases. These methods are favored for their low cost and minimal participant burden. However, a growing body of evidence demonstrates that these approaches constitute a significant methodological weakness, introducing substantial error and misclassification that can compromise data integrity and lead to erroneous conclusions. This application note synthesizes current evidence to delineate the specific limitations of traditional methods and provides standardized, validated protocols to enhance the rigor and reproducibility of menstrual cycle research.

Quantitative Evidence of Methodological Inaccuracy

Empirical studies consistently reveal poor agreement between self-reported menstrual cycle data and prospectively measured gold standards. The following tables summarize key quantitative findings on the inaccuracy of self-reporting and the failure of calendar-based methods to correctly identify key hormonal events.

Table 1: Documented Inaccuracy of Self-Reported Menstrual Cycle Length

Study Finding	Quantitative Data	Citation
Misclassification of Cycle Length Category	21% of women were misclassified when self-reported cycle length was categorized (<26, 26-35, >35 days).	[1]
Substantial Measurement Error	43% of women self-reported a "usual" cycle length that was >2 days different from their prospectively measured mean length.	[1]
Systematic Overestimation	On average, women overestimated their cycle length by 0.7 days (95% CI: 0.3, 1.0).	[2]
Discrepancy with Hormone-Monitored Data	Calculated cycle lengths from quantitative hormone monitoring were frequently shorter than user-reported cycle lengths.	[3]

Table 2: Failure of Calendar-Based Methods to Identify Hormonal Phase

Calendar-Based Method	Criterion for Ovulation (Progesterone >2 ng/mL)	Criterion for Mid-Luteal Phase (Progesterone >4.5 ng/mL)	Citation
Counting forward 10-14 days from menses	18% of women met the criterion	Information missing	[4]
Counting back 12-14 days from cycle end	59% of women met the criterion	Information missing	[4]
Using a positive urinary ovulation test + 1-3 days	76% of women met the criterion	58-75% of women met the criterion (7-9 days post-ovulation test)	[4]

The data in Table 2 underscore that assumptions about phase timing are often incorrect. Relying on a fixed 28-day model is flawed, as only a small fraction of individuals ovulate on cycle day 14, even among those with regular cycles [3]. Furthermore, self-report cannot detect subtle menstrual disturbances like anovulatory or luteal phase deficient cycles, which are present in up to 66% of exercising females and present meaningfully different hormonal profiles [5].

Experimental Validation Protocols

To address the limitations of traditional methods, researchers should adopt the following validated experimental protocols for accurate cycle phase determination.

Protocol for Urinary Hormone Monitoring to Confirm Ovulation and Luteal Phase

Objective: To prospectively identify the fertile window, confirm ovulation, and verify the luteal phase using at-home urinary hormone tests.

Background: The luteinizing hormone (LH) surge is a definitive precursor to ovulation. A subsequent rise in progesterone (measured via its urinary metabolite, pregnanediol glucuronide, PdG) confirms that ovulation has occurred [3] [4].

Materials:

• Quantitative urinary LH and PdG test kits (e.g., Oova, Mira, Initio).
• Smartphone application for reading test strips and tracking data.
• Laboratory equipment for serum progesterone validation (optional).

Procedure:

• Initiation: Participants begin testing on day 8 of their cycle (first day of menses = day 1).
• Testing Frequency: Test urine at the same time each day until a positive ovulation test is identified.
• Identification of LH Surge: The LH peak is identified when levels rise significantly above the participant's established baseline [3].
• Confirmation of Ovulation: Continue testing for 3-7 days after the LH peak to detect a sustained rise in PdG levels. A PdG level >5 μg/mL is often used to confirm ovulation.
• Luteal Phase Assessment: To capture the mid-luteal phase, schedule laboratory visits or additional assessments for 7-9 days after a positive ovulation test [4].

Validation: This method was validated in a study of 1,233 users (4,123 cycles), which successfully identified ovulation and detailed phase-length variability across age groups [3].

Protocol for Serum Hormone Verification of Menstrual Cycle Phase

Objective: To provide definitive, retrospective confirmation of menstrual cycle phase through the analysis of serum sex steroid hormone concentrations.

Background: While resource-intensive, serum hormone analysis remains the gold standard for validating cycle phase and detecting subtle endocrine disturbances [5] [4].

Materials:

• Phlebotomy supplies.
• Laboratory capable of performing radioimmunoassay (RIA) or electrochemiluminescence immunoassay (ECLIA) for estradiol (E2) and progesterone (P4).
• Centrifuge and freezer for serum storage.

Procedure:

• Blood Collection: Collect blood samples in the morning to control for diurnal variation.
• Phase-Specific Sampling:
  • Early Follicular Phase: Sample on days 2-5 of the cycle. Expected hormone levels: low E2 and P4.
  • Peri-Ovulatory Phase: Sample based on a detected LH surge or when cervical mucus indicates fertility. Expected hormone levels: high E2, low P4.
  • Mid-Luteal Phase: Sample approximately 7 days after a confirmed ovulation. Expected hormone levels: elevated P4 (>4.5 ng/mL is indicative of the mid-luteal phase) and secondary E2 peak [4].
• Hormone Assay: Analyze serum for E2 and P4 concentrations.
• Cycle Validation: Confirm an ovulatory cycle by the presence of a luteal phase P4 level >2 ng/mL [4]. Luteal phase deficiency may be suspected if P4 levels are consistently low.

Visualizing the Research Workflow and Hormonal Signaling

The following diagrams illustrate the standardized workflow for accurate cycle phase determination and the underlying hormonal pathways that these methods track.

Figure 1: Workflow for Urinary Hormone Monitoring to Confirm Cycle Phase.

Figure 2: Simplified Hypothalamic-Pituitary-Ovarian (HPO) Axis Signaling.

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Key Research Reagent Solutions for Menstrual Cycle Phase Determination

Item	Function/Application in Research
Urinary Luteinizing Hormone (LH) Tests	Predicts ovulation by detecting the LH surge in urine. Essential for pinpointing the start of the fertile window and peri-ovulatory phase.
Urinary PdG (Pregnanediol Glucuronide) Tests	Confirms ovulation by measuring the urinary metabolite of progesterone. A sustained rise indicates a viable luteal phase.
Automated Basal Body Temperature (BBT) Devices	Detects the slight, sustained rise in resting body temperature that follows ovulation. Useful for retrospective confirmation of ovulation.
Serum Progesterone Immunoassay	The gold-standard method for quantifying serum progesterone to definitively confirm ovulation and assess luteal function.
Structured Menstrual Cycle Diary	Prospective daily record of bleeding and symptoms. Provides foundational data for calculating cycle length and identifying patterns when used with hormone data.
Data Analysis Software (R, SAS)	For implementing multilevel statistical models that account for within-person hormone variance across the cycle, as recommended by best practices [6].

The limitations of traditional counting and self-report methods present a significant challenge to the validity of menstrual cycle research. Quantitative evidence clearly demonstrates that these approaches lead to high rates of misclassification and systematic error. To ensure the generation of robust, reliable, and reproducible data—particularly in critical fields like drug development and clinical trial design—researchers must transition to methodologically superior protocols. The integration of quantitative urinary hormone monitoring and strategic serum validation, as outlined in this application note, provides a feasible and scientifically rigorous path forward. Adopting these standardized tools will empower researchers to accurately account for the menstrual cycle as a biological variable, thereby enhancing the precision of scientific discoveries and the efficacy of future therapeutics.

{ARTICLE CONTENT ENDS HERE}

The High Prevalence of Undetected Menstrual Disturbances in Study Populations

Accurate characterization of the menstrual cycle is fundamental to reproductive health research, yet methodological inconsistencies often obscure the true prevalence of menstrual disturbances. Evidence confirms that a significant proportion of individuals experience abnormal uterine bleeding (AUB), with one in three women reporting excessive menstrual loss, a figure rising to one in two as menopause approaches [7]. Furthermore, emerging data indicates that conditions like long COVID are associated with increased menstrual disturbances, including heavier bleeding, prolonged duration, and intermenstrual bleeding [7]. This application note details standardized protocols for coding menstrual cycle phase and detecting menstrual disturbances, providing researchers with tools to minimize classification error and enhance data validity in clinical and research settings. Establishing methodological rigor is essential for generating reproducible findings that accurately reflect underlying female physiology.

Quantifying the Scope of the Problem

Prevalence of Menstrual Disturbances

Recent large-scale studies reveal that menstrual disturbances are highly prevalent yet frequently under-detected in research populations due to inconsistent assessment methods. The table below summarizes key prevalence data from recent investigations.

Table 1: Documented Prevalence of Menstrual Disturbances in Research Populations

Study Population	Disturbance Type	Prevalence	Citation
General Population (Pre-pandemic)	Heavy Menstrual Bleeding (HMB)	1 in 3 women (rising to 1 in 2 pre-menopause)	[7]
UK COVID-19 Survey (n=12,187)	Any abnormal menstrual symptom at baseline	57% of participants	[7]
UK COVID-19 Survey: Long COVID Group (n=1,048)	Increased menstrual volume, duration, and intermenstrual bleeding vs. never-infected	Significantly increased	[7]
Long COVID Patients (Patient-led survey)	Any menstrual issues	33.8% reported	[7]
Working U.S. Females (n=372)	Menstrual pain (dysmenorrhea)	Up to 91% (29% severe pain)	[8]
Women with Menstrual Irregularities (n=150, India)	Co-occurring hypothyroidism	24% of participants	[9]
Hypothyroid Women	Any menstrual irregularity	23.4% (vs. 12% in euthyroid controls)	[9]

Impact on Research Outcomes

Undetected menstrual disturbances introduce significant confounding variability that can compromise research integrity and drug development outcomes. Menstrual symptoms have demonstrated substantial impact on functional outcomes, including work-related productivity [8]. In the U.S., annual indirect costs of menstrual bleeding disorders were estimated at $12 billion, while productivity loss from menstrual-related symptoms costs employers $225.8 billion annually [7] [8]. This economic burden underscores the critical need for precise detection and classification in clinical trials.

Furthermore, hormonal fluctuations across the cycle can modulate other disease states. Over one-third of menstruating patients with long COVID report exacerbation of their symptoms the week before or during menses [7]. Failure to account for these cyclic patterns can lead to inaccurate assessment of treatment efficacy and side effect profiles in drug development.

Experimental Protocols for Menstrual Cycle Phase Determination

Gold-Standard Hormonal Assessment

Objective: To determine menstrual cycle phase through direct measurement of circulating ovarian hormones. Background: Menstrual cycle phases are defined by specific hormonal milieus. The follicular phase is characterized by low progesterone and rising estradiol, while the luteal phase features sustained high progesterone following ovulation [6] [10].

Table 2: Essential Research Reagents for Hormonal Phase Determination

Reagent/Instrument	Specification	Function
Chemiluminescence Immunoassay System	e.g., Roche Cobas e411 analyzer	Quantifies serum/plasma hormone concentrations with high sensitivity
Estradiol (E2) Assay Kit	Serum/Plasma/Saliva format	Measures estradiol levels for follicular phase and ovulation identification
Progesterone (P4) Assay Kit	Serum/Plasma/Saliva format	Confirms ovulation and identifies luteal phase
Luteinizing Hormone (LH) Assay Kit	Urine/Serum format	Detects LH surge predicting ovulation
Thyroid Function Test Panel	TSH, free T4, free T3	Rules out thyroid dysfunction as cause of menstrual irregularity [9]

Procedure:

• Sample Collection: Collect fasting venous blood samples (5 mL) in the morning. For multi-phase studies, collect samples at a minimum of three timepoints across one cycle [6].
• Hormone Analysis: Process samples using chemiluminescence immunoassay per manufacturer protocols. Standard reference ranges:
  • Follicular Phase: Low P4 (<1.5 ng/mL), variable E2
  • Ovulation: E2 >200 pg/mL, LH surge >20 IU/L
  • Mid-Luteal Phase: P4 >5 ng/mL, E2 100-200 pg/mL [10]
• Phase Confirmation: Classify phase using both absolute hormone levels and within-person hormone changes across assessments. Avoid relying solely on population-based hormone ranges, which have poor accuracy for individual phase determination [10].

Protocol for Prospective Symptom and Cycle Tracking

Objective: To characterize menstrual cycle patterns and identify disturbances through prospective daily monitoring. Background: Retrospective recall of menstrual symptoms shows poor agreement with prospective daily ratings, with a notable bias toward false positive reports [6].

Procedure:

• Participant Selection: Recruit naturally-cycling individuals. Exclude those pregnant, breastfeeding, perimenopausal, or using hormonal contraception [8].
• Cycle Tracking: Participants record daily via diary or mobile application:
  • Menstrual bleeding start/end dates and intensity
  • Core symptoms (e.g., pain, mood changes, energy)
  • Basal body temperature (BBT) upon waking
  • Urinary LH surge testing if fertility tracking is needed
• Duration: Minimum two consecutive menstrual cycles to establish pattern and identify irregularities [6].
• Data Analysis: Apply standardized criteria like the Carolina Premenstrual Assessment Scoring System (C-PASS) for premenstrual disorders. Code cycle phases after the fact using confirmed menstrual dates [6].

Emerging Protocol: Machine Learning with Wearable Device Data

Objective: To classify menstrual cycle phases using physiological signals from wearable devices. Background: Recent advances in wearable technology and machine learning enable passive, continuous cycle phase detection using physiological signals like skin temperature, heart rate, and heart rate variability [11].

Procedure:

• Data Collection: Participants wear a validated wrist-worn device (e.g., Empatica E4, Oura Ring) continuously for multiple cycles, collecting:
  • Skin temperature
  • Heart rate (HR) and interbeat interval (IBI)
  • Electrodermal activity (EDA)
  • Sleep metrics
• Ground Truth Labeling: Establish phase labels using urinary LH tests for ovulation and self-reported monset dates.
• Feature Engineering: Extract daily features from raw signals. Use fixed windows (e.g., 7-day non-overlapping) or rolling windows for daily phase prediction.
• Model Training: Train a Random Forest or other classifier using a leave-last-cycle-out or leave-one-subject-out cross-validation approach.
• Validation: Compare model predictions against ground truth labels. Reported accuracies: 87% for 3-phase classification (period, ovulation, luteal) and 68% for 4-phase classification [11].

Data Analysis and Coding Recommendations

Statistical Modeling for Menstrual Cycle Data

The menstrual cycle is fundamentally a within-person process and should be analyzed using appropriate statistical methods that account for this nested structure [6] [12].

Recommended Approaches:

• Multilevel Modeling (MLM): Treat repeated measurements (level 1) as nested within persons (level 2). This separates within-person variance from between-person variance.
• Minimal Sampling: Collect at least three observations per person across one cycle to estimate random effects. For greater reliability of between-person differences in within-person changes, collect three or more observations across two cycles [6].
• Cycle Day Coding: Calculate cycle day using a combination of forward-count (from prior monset) and backward-count (to next monset) methods based on confirmed cycle dates [12].
• Data Visualization: Create spaghetti plots of raw and person-centered outcomes for each participant individually before group-level analysis to detect outliers and patterns [12].

Common Methodological Pitfalls and Solutions

Table 3: Common Methodological Errors in Menstrual Cycle Research and Recommended Solutions

Methodological Error	Impact on Data Quality	Recommended Solution
Reliance on retrospective recall of cycle dates or symptoms	High rate of false positives; poor agreement with prospective measures [6]	Implement prospective daily monitoring for minimum 2 cycles
Using between-subjects designs for cycle effects	Conflates within- and between-person variance; invalid conclusions [6]	Use repeated-measures designs with appropriate multilevel modeling
Phase determination by counting forward from menses only	High error rate due to cycle length variability [10]	Use backward counting from next menses or hormonal confirmation
Defining phase using population hormone ranges	Poor accuracy for individual classification [10]	Use within-person hormone changes or combined methods
Failure to screen for premenstrual disorders	Confounds general cycle effects with pathological responses [6]	Screen for PMDD/PME using validated tools like C-PASS
Ignoring thyroid dysfunction in irregular cycles	Misses underlying etiology of menstrual disturbance [9]	Include thyroid function tests (TSH, free T4) in screening

The high prevalence of undetected menstrual disturbances represents a critical methodological challenge that can significantly compromise research validity and drug development outcomes. By implementing the standardized protocols detailed in this application note—including gold-standard hormonal assessment, prospective symptom tracking, and emerging wearable technology approaches—researchers can significantly improve the accuracy of menstrual cycle phase determination. Adopting these rigorous methodologies will enhance detection of true menstrual disturbances, facilitate more precise investigation of cycle-mediated disease states, and ultimately strengthen the evidence base for women's health interventions. The field must move beyond error-prone retrospective methods and embrace validated, prospective approaches that account for the substantial within-person and between-person variability inherent in menstrual cycle physiology.

The menstrual cycle is a quintessential example of a complex, dynamic endocrine system, governed by precise fluctuations of key reproductive hormones. For researchers and drug development professionals, the accurate quantification of these hormones is critical for diagnosing disorders, developing therapeutics, and advancing personalized medicine. Biosensing technologies have emerged as powerful tools to meet this need, enabling the precise, rapid, and often non-invasive measurement of hormonal biomarkers. This document details the physiological basis of these hormonal fluctuations, the biosensing principles used for their measurement, and standardized protocols for their application in research settings, with a specific focus on coding phase determination protocols.

Physiological Background: The Menstrual Cycle and Key Hormones

The human menstrual cycle is a biphasic process, typically lasting between 24 to 38 days, and is orchestrated by the hypothalamic-pituitary-ovarian (HPO) axis. It is characterized by coordinated feedback loops that result in distinct phases: the follicular phase (including menses and ending with ovulation) and the luteal phase [11] [13]. These phases are defined by critical biological events—ovum development, ovulation, and preparation for potential implantation—driven by the following key hormones:

• Follicle-Stimulating Hormone (FSH): Produced by the pituitary gland, FSH stimulates the growth and maturation of ovarian follicles in the early follicular phase.
• Luteinizing Hormone (LH): Also secreted by the pituitary, a surge in LH triggers ovulation, the release of a mature oocyte from the dominant follicle.
• Estrogen (primarily measured as Estrone-3-glucuronide, E1-3G, in urine): Produced by the developing follicles, estrogen levels rise through the follicular phase, peaking just before ovulation to trigger the LH surge. It is responsible for proliferating the uterine endometrium.
• Progesterone (primarily measured as Pregnanediol Glucuronide, PDG, in urine): Secreted by the corpus luteum after ovulation, progesterone levels rise during the luteal phase to prepare the uterine lining for implantation and support early pregnancy.

The following diagram illustrates the logical relationships and feedback mechanisms within the HPO axis that govern the menstrual cycle, providing a framework for understanding measurable hormonal patterns.

Analytical Techniques for Hormone Detection

A variety of biosensing platforms have been developed to detect and quantify reproductive hormones, each with distinct operational principles and analytical merits. The choice of technique depends on the required sensitivity, specificity, form factor (e.g., lab-based vs. point-of-care), and the sample matrix (e.g., serum, urine, saliva) [14].

Table 1: Comparison of Hormone Biosensing Techniques

Technique	Principle	Bioreceptor Examples	Typical Sample	Key Advantages	Reported Limitations
Electrochemical (EC)	Measures electrical current/potential change upon hormone-bioreceptor binding.	Antibodies, Aptamers, Enzymes [14]	Serum, Urine	High sensitivity, portability, cost-effectiveness [14]	Potential interference in complex matrices.
Optical	Detects changes in light properties (absorbance, fluorescence, SPR).	Antibodies, DNA [14]	Serum, Urine	High specificity and multiplexing potential.	Instrumentation can be complex and expensive.
Photoelectrochemical (PEC)	Uses light to excite a photosensitizer; measures resulting photocurrent.	Antibodies, Peptides [14]	Serum, Urine	Very low background noise, high sensitivity [14]	Relative novelty; requires sophisticated material design.
Electrochemiluminescence (ECL)	Generates light from electrochemical reactions at an electrode surface.	Antibodies, Aptamers [14]	Serum, Urine	Excellent sensitivity and wide dynamic range [14]	Requires specific reagents (e.g., ruthenium complexes).
Lateral Flow Immunoassay (LFIA)	Capillary action moves sample over immobilized antibodies; visual readout.	Antibodies (e.g., anti-LH) [15]	Urine	Rapid, low-cost, ideal for home-use (e.g., ovulation predictors).	Semi-quantitative at best, limited dynamic range.

Experimental Protocols for Hormone Monitoring and Phase Determination

This section outlines detailed protocols for two primary approaches to menstrual cycle monitoring: a gold-standard laboratory validation method and an emerging approach using wearable sensor data.

Protocol 4.1: Gold-Standard Quantitative Menstrual Cycle Monitoring with Urine Hormones and Ultrasound Validation

This protocol, adapted from established research methodologies, aims to characterize urinary hormone patterns and validate them against the clinical gold standards of serum hormone levels and transvaginal ultrasound for ovulation confirmation [13].

Objective: To establish a quantitative correlation between urine hormone concentrations (FSH, E1-3G, LH, PDG) measured by a biosensor (e.g., Mira monitor) and the day of ovulation confirmed by ultrasound in participants with regular and irregular cycles.

Materials & Reagents:

• Quantitative Urine Hormone Monitor (e.g., Mira monitor) and corresponding test strips [13].
• Serum Hormone Assay Kits for FSH, LH, Estradiol, and Progesterone (e.g., CLIA or ELISA).
• Ultrasound Machine with endovaginal transducer.
• Data Collection App for tracking bleeding patterns and symptoms [13].

Procedure:

• Participant Recruitment & Screening: Recruit three cohorts using purposive sampling: Group 1 (regular cycles, 24-38 days), Group 2 (irregular cycles with PCOS), Group 3 (irregular cycles, athletes). Obtain ethical approval and informed consent [13].
• Longitudinal Sample & Data Collection:
  • Participants collect first-morning urine samples daily for one complete menstrual cycle (or 3 months for robust data).
  • Urine is analyzed immediately using the quantitative hormone monitor.
  • Serum samples are collected twice per week during the follicular phase and daily around the expected LH surge for correlation with urine data.
• Ultrasound Confirmation of Ovulation:
  • Perform serial transvaginal ultrasounds every 1-3 days starting around cycle day 10.
  • Track the growth of the dominant follicle until it disappears post-ovulation. The day of ovulation is defined as the day prior to the follicle's disappearance [13].
• Data Integration and Analysis:
  • Align urine hormone profiles (LH peak, PDG rise) with the ultrasound-defined day of ovulation (Day 0).
  • Perform statistical correlation (e.g., Pearson's correlation) between quantitative urine hormone values and matched serum hormone concentrations.
  • Establish normative hormone patterns for regular cycles and compare them to patterns observed in PCOS and athletic cohorts.

Protocol 4.2: Menstrual Phase Identification Using Machine Learning and Wearable Device Data

This protocol leverages continuous physiological data from wearable devices to classify menstrual cycle phases using machine learning models, reducing the burden of self-reporting [11].

Objective: To train and validate a machine learning model (e.g., Random Forest) to identify menstrual cycle phases (Menses, Follicular, Ovulation, Luteal) based on physiological signals from a wrist-worn device.

Materials & Reagents:

• Wrist-worn Wearable Device capable of measuring: Skin Temperature, Heart Rate (HR), Interbeat Interval (IBI)/Heart Rate Variability (HRV), and Electrodermal Activity (EDA) [11].
• Urine LH Test Kits (for labeling the ovulation phase in the training data).
• Data Processing Software (e.g., Python with scikit-learn, Pandas, NumPy).

Procedure:

• Data Collection and Labeling:
  • Recruit participants to wear the device continuously for 2-5 months.
  • Collect daily self-reported menses onset.
  • Use urine LH test kits to identify the LH surge. Define the ovulation phase as the period spanning 2 days before to 3 days after the positive LH test [11].
  • Label the data into four phases based on the reference definitions:
    • Menses (M): Days of menstrual bleeding.
    • Follicular (F): Post-menses until the start of the ovulation window.
    • Ovulation (O): LH surge window (± 2 days).
    • Luteal (L): Post-ovulation until next menses.
• Feature Extraction:
  • From the raw signals (Temperature, HR, IBI, EDA), extract statistical features (e.g., mean, standard deviation, min, max) over 24-hour non-overlapping windows.
  • For daily prediction, use a rolling window (e.g., 24-48 hours) to extract features.
• Model Training and Validation:
  • Use a Leave-Last-Cycle-Out cross-validation: train on a participant's first n-1 cycles and test on their last unseen cycle [11].
  • Train multiple classifiers (e.g., Random Forest, Logistic Regression, Support Vector Machines).
  • Evaluate performance using accuracy, precision, recall, F1-score, and Area Under the ROC Curve (AUC-ROC).

Reported Performance: Using this methodology with a Random Forest classifier and a fixed window for three phases (M, O, L), an accuracy of 87% and an AUC-ROC of 0.96 have been achieved [11].

The workflow for this machine learning-based protocol is summarized below.

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials for Hormonal Biosensing Research

Item	Function & Application in Research	Example Use Case
Quantitative Urine Hormone Monitor (e.g., Mira, Inito)	Provides numerical concentration values for FSH, E1-3G, LH, PDG in urine. Essential for establishing detailed hormone profiles and correlating with other biomarkers [13].	Gold-standard protocol for predicting and confirming ovulation [13].
Lateral Flow Ovulation Tests (qualitative LH strips)	Detects the presence of LH above a threshold to identify the LH surge. Low-cost method for generating ground-truth labels for the ovulation phase [11] [15].	Labeling the "Ovulation" class in machine learning model training datasets [11].
Basal Body Temperature (BBT) Thermometer	Measures subtle, progesterone-mediated temperature shifts post-ovulation. A traditional symptomothermal method to confirm ovulation has occurred [15].	Used in FEMM and sympto-thermal methods to cross-validate ovulation [15].
Electrochemical Sensor Strips	The transducer component in many biosensors. Can be functionalized with specific bioreceptors (e.g., anti-LH antibodies) for hormone detection in lab-developed tests [14].	Developing novel, low-cost, point-of-care biosensors for hormone detection in research settings [14].
Specific Bioreceptors (Antibodies, Aptamers)	Provides the molecular recognition element for a biosensor, ensuring high specificity for the target hormone (e.g., LH, Progesterone) [14].	Immobilization on sensor surfaces (e.g., electrodes, SPR chips) to create the core sensing interface.

Data Presentation and Analysis

Table 3: Quantitative Performance of Menstrual Phase Identification Methods

Method	Phases Classified	Key Metrics	Performance & Notes	Source
Machine Learning (Random Forest) on Wearable Data	3 Phases: Menses, Ovulation, Luteal	Accuracy: 87%AUC-ROC: 0.96	Fixed window feature extraction. Leave-last-cycle-out validation. High performance for 3-class problem.	[11]
Machine Learning (Random Forest) on Wearable Data	4 Phases: Menses, Follicular, Ovulation, Luteal	Accuracy: 71%AUC-ROC: 0.89	Fixed window feature extraction. Performance decreases with finer phase granularity.	[11]
Machine Learning (Random Forest) - Daily Tracking	4 Phases	Accuracy: 68%AUC-ROC: 0.77	Rolling window feature extraction. Reflects challenge of real-time, daily phase estimation.	[11]
In-ear Wearable Temperature Sensor	Ovulation Occurrence	Accuracy: 76.9%	Identified ovulation in 30/39 cycles using a Hidden Markov Model on continuous temperature data.	[11]
Marquette Method (Urine Hormone Monitor + Algorithm)	Fertile Window	N/A	Uses Clearblue or Mira monitor with a specific protocol. Reported as highly effective for family planning, especially in postpartum.	[16] [15]

In menstrual cycle research, the fundamental distinction between fertile window prediction and broad phase classification dictates every subsequent choice in study design, measurement technology, and analytical methodology. Fertile window prediction focuses precisely on identifying the brief period encompassing ovulation and the days prior when conception is possible, requiring high temporal resolution and precision [17]. In contrast, broad phase classification categorizes the cycle into larger physiological phases—typically the menstrual, follicular, ovulatory, and luteal phases—to investigate longer-duration hormonal effects on physiological, cognitive, or performance outcomes [6] [11]. The protocol objective must be clearly defined at the outset, as it determines the requisite measurement frequency, technology selection, and validation protocols. This document provides a structured framework for selecting and implementing these distinct methodological approaches within a research context.

Performance Comparison of Tracking Objectives

The choice between protocol objectives is guided by their differing performance characteristics, accuracy, and suitability for various research endpoints. The table below summarizes the primary focuses and validated performance metrics for the two approaches.

Table 1: Core Objective Comparison for Menstrual Cycle Tracking Protocols

Protocol Objective	Primary Research Focus	Key Performance Metrics (from literature)	Optimal Use Cases
Fertile Window Prediction	Pinpointing ovulation and the days of peak fertility.	- Wearable Physiology (Oura Ring): 96.4% detection rate; MAE: 1.26 days [17].- Calendar Method: MAE: 3.44 days [17].- Cervical Mucus Tracking: 48-76% accuracy within 1 day [17].	Fertility studies (conception/contraception), precise hormonal event correlation.
Broad Phase Classification	Categorizing cycle into multi-day phases (e.g., Follicular, Ovulatory, Luteal).	- Machine Learning (Wristwear): Up to 87% accuracy (3-phase) [11].- Machine Learning (Sleep HR): Improved luteal phase recall [18].- BBT: Limited by sleep timing variability [18].	Investigating cycle-phase effects on symptoms, performance, sleep, or mood.

Performance data reveals a significant accuracy advantage for physiology-based methods over calendar-based counting for both objectives [18] [17]. The most appropriate method depends on the required precision and the specific biological process under investigation.

Detailed Experimental Protocols

Protocol for Fertile Window Prediction and Ovulation Detection

This protocol is designed for studies requiring precise identification of the fertile window, such as those investigating fertility, hormonal contraception efficacy, or the acute effects of peri-ovulatory hormonal shifts.

Objective: To accurately identify the day of ovulation and the preceding fertile window in a natural menstrual cycle.

Materials:

• Primary (Reference) Method: Urinary Luteinizing Hormone (LH) test kits (e.g., lateral flow assays).
• Experimental Method(s): Wearable sensor(s) capable of continuous physiological monitoring (e.g., Oura Ring for distal body temperature, heart rate; or validated wrist-worn devices).
• Data Collection Platform: Customized mobile application for self-reporting.

Procedure:

• Participant Screening & Consent: Recruit participants with ovulatory cycles (historical cycle length 21-35 days). Exclude those using hormonal contraception, pregnant, breastfeeding, or with self-reported conditions like PCOS.
• Baseline Data Collection:
  • Record demographic information, health history, and typical cycle characteristics.
  • Instruct participants on the proper use of all devices and self-reporting tools.
• Longitudinal Data Acquisition:
  • Urinary LH Testing: Participants begin daily testing from cycle day 10 until a positive surge is detected. The reference ovulation date is defined as the day after the last positive LH test [17].
  • Wearable Data: Participants wear the designated sensor continuously, preferably during sleep, to collect physiology data (e.g., skin temperature, heart rate).
  • Cycle Tracking: Participants log the first day of menstruation (cycle day 1) and the start of subsequent cycles in the app.
• Data Processing & Algorithm Application:
  • For Wearable Physiology: Process raw sensor data (e.g., filter, impute missing values) and apply a validated algorithm to detect the sustained temperature rise that confirms ovulation has occurred [17].
  • For Calendar Method (Benchmark): Calculate estimated ovulation day as: Median Cycle Length (last 6 months) - 14 days [17].
• Validation & Analysis:
  • Compare the ovulation day estimated by the experimental method(s) against the reference LH test date.
  • Calculate performance metrics: detection rate, mean absolute error (MAE), and mean error (bias).

Protocol for Broad Menstrual Phase Classification

This protocol is suited for research examining how longer-term hormonal states influence outcomes like athletic performance, cognitive function, sleep quality, or mood disorders.

Objective: To classify the menstrual cycle into distinct, hormonally-defined phases (e.g., Follicular, Ovulatory, Luteal) for investigating phase-dependent effects.

Materials:

• Phase Determination Tools:
  • Gold Standard: Quantitative urine hormone monitor (e.g., Mira monitor) tracking E1G, LH, and PdG [13].
  • Acceptable Alternative: Combination of forward/backward counting from menses with salivary hormone (estradiol, progesterone) sampling twice weekly [19].
  • Experimental Method: Multi-sensor wearable device (e.g., wrist-worn device measuring skin temperature, HR, HRV, EDA) [11].
• Outcome Measures: Questionnaires, cognitive tests, or performance metrics relevant to the study.

Procedure:

• Participant Screening & Consent: As in Protocol 3.1, with emphasis on recruiting participants whose characteristics match the research question (e.g., athletes, individuals with PMDD).
• Baseline & Daily Data:
  • Record baseline data and the first day of menses (cycle day 1).
  • Participants complete daily outcome measures (e.g., symptom logs).
• Phase Determination via Gold Standard:
  • Using Urine Hormones: Participants use the monitor daily. Phases are defined as:
    • Menstrual: Cycle days 1-5 (bleeding).
    • Follicular: From menses end until LH surge.
    • Ovulation: LH peak ± 3 days [11].
    • Luteal: From ovulation confirmation (via rising PdG) until next menses [13].
• Experimental Data Collection:
  • Collect continuous physiological data from the wearable device.
  • Schedule outcome assessments (e.g., lab visits) based on the phase determined by the gold standard method.
• Machine Learning for Phase Classification:
  • Extract features (e.g., mean nocturnal HR, skin temperature, HRV) from the wearable data within each gold-standard-labeled phase.
  • Train a classifier (e.g., Random Forest) using a leave-one-subject-out or leave-last-cycle-out cross-validation approach to predict phase labels [11].
• Validation & Analysis:
  • Assess classifier performance (accuracy, precision, recall, F1-score) against the gold-standard phase labels.
  • Statistically model the association between the classified phases and the outcome measures of interest.

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 2: Key Materials and Reagents for Menstrual Cycle Research Protocols

Item	Function & Application in Protocol
Urinary LH Test Kits	Detects the luteinizing hormone surge, providing a standard reference for estimating ovulation day [17].
Quantitative Urine Hormone Monitor (e.g., Mira)	Measures multiple hormones (E1G, LH, PdG) to enable precise, at-home phase classification and ovulation confirmation [13].
Salivary Hormone Immunoassay Kits	Allows non-invasive, frequent sampling of estradiol and progesterone levels for phase verification [19].
Wearable Sensors (Oura Ring, E4/EmbracePlus Bands)	Collects continuous physiological data (skin temperature, HR, HRV, EDA) for algorithm-based ovulation detection and phase classification [18] [17] [11].
Custom Mobile Application	Platform for participants to self-report menses, symptoms, and LH test results; facilitates data integration [13].

Workflow Visualization

Figure 1. Decision Workflow for Menstrual Cycle Research Objectives. This diagram outlines the critical decision points and methodological pathways for research focused on Fertile Window Prediction (green) versus Broad Phase Classification (blue). The choice of initial objective directly determines the appropriate gold standard tools, experimental data streams, and validation methodologies.

Methodologies in Practice: From Biosignal Acquisition to Machine Learning Models

Accurately determining menstrual cycle phase is critical for research on female physiology, psychology, and drug development. Traditional methods for phase determination have relied heavily on self-reported data, which often introduces significant error [10]. Emerging technologies now enable more precise, continuous, and objective tracking of cycle phases through various data sources, including consumer wearables, research-grade sensors, and novel contactless monitoring systems. These technologies leverage physiological parameters such as heart rate, skin temperature, and respiratory rate that fluctuate predictably across the menstrual cycle in response to hormonal changes [20] [21]. This document provides a comprehensive overview of available data sources, their experimental protocols, and integration frameworks for menstrual cycle research.

The table below summarizes the key characteristics, measured parameters, and performance metrics of different data sources used for menstrual cycle phase tracking.

Table 1: Comparison of Menstrual Cycle Tracking Data Sources

Data Source	Key Measured Parameters	Reported Accuracy/Performance	Key Advantages	Key Limitations
Wrist-worn Wearables (e.g., Ava Bracelet, Oura Ring)	Resting Heart Rate, Heart Rate Variability (HRV), Wrist Skin Temperature (WST), Respiratory Rate, Skin Perfusion [20] [21]	90% accuracy in detecting fertile window (Ava) [21]; 87% accuracy in 3-phase classification (Random Forest model) [11]	Continuous, multi-parameter monitoring under free-living conditions; Rich data for machine learning [20] [11]	Consumer devices not always validated for research; Potential privacy concerns with data [20]
Intravaginal Sensors (e.g., OvulaRing)	Core Body Temperature [20] [22]	99% accuracy in detecting ovulation; 89% accuracy in predicting ovulation [22]	High accuracy for ovulation; Measures core body temperature directly [22]	Invasive form factor; Lower user acceptability and comfort for long-term use [22]
Contactless/Non-Invasive Sensors (e.g., Radar, LiDAR, PPG)	Heart Rate, Vascular Activity, Breathing Rhythm [23]	Framework proposed; Specific accuracy data under evaluation [23]	Privacy-preserving; No skin contact required; Suitable for sensitive populations [23]	Emerging technology; Requires further clinical validation [23]
Manual Input/Traditional Methods (e.g., BBT, Calendar Apps)	Basal Body Temperature (BBT), Menstrual Start Date, Urinary Luteinizing Hormone (LH) [10] [6]	Error-prone; Self-report projection methods resulted in phase misclassification [10]	Low cost; Accessible without specialized hardware [10]	Susceptible to user error and recall bias; Poor predictor of fertile window prospectively [21] [10]

Experimental Protocols for Key Tracking Modalities

Protocol for Multi-Parameter Wearable Data Collection (e.g., Wrist-worn Sensors)

Objective: To collect high-frequency physiological data under free-living conditions for menstrual cycle phase classification and ovulation prediction [21] [11].

Materials:

• Validated wearable sensor (e.g., research-grade wristband, Oura Ring, Ava Bracelet)
• Smartphone application for data syncing
• Urinary luteinizing hormone (LH) tests (e.g., Clearblue Digital Ovulation Test) for ground truth ovulation labeling [21] [11]
• Electronic diary for logging activities, symptoms, and confounders (e.g., alcohol, caffeine, exercise, large meals) [21] [24]

Procedure:

• Participant Screening: Recruit naturally-cycling females meeting inclusion criteria (e.g., age 18-40, regular cycles, no hormonal contraception, not pregnant or lactating) [21] [6].
• Device Fitting and Calibration: Instruct participants on proper device wear (e.g., snug fit on wrist, continuous overnight wear) [21].
• Data Collection Period: Participants wear the device nightly during sleep for the duration of the study (e.g., up to 12 months or multiple cycles) [21].
• Ground Truth Labeling:
  • Ovulation Confirmation: Perform urinary LH tests daily from cycle day 6 until a surge is detected to pinpoint the end of the fertile window [21] [11].
  • Menstrual Logging: Record first day of menstruation (Cycle Day 1) and subsequent bleeding days via app or diary [6].
• Data Preprocessing: Extract nightly averages or minute-level data for parameters: skin temperature, heart rate, HRV, respiratory rate [21] [11].
• Confounder Adjustment: Statistically control for factors like alcohol consumption, intense evening exercise, and large meals before sleep, which can affect physiological readings [24].

Protocol for Contactless Menstrual Health Prediction

Objective: To leverage contactless biosensing and federated learning for privacy-preserving menstrual health prediction [23].

Materials:

• Multi-modal sensor system integrating radar-based respiration sensing, photoplethysmography (PPG), and LiDAR-assisted microvascular mapping [23].
• Edge computing device (e.g., smartphone or dedicated hardware) for local data processing.
• Central server for federated learning model aggregation.

Procedure:

• Signal Acquisition: Set up contactless sensors in the participant's sleeping environment to continuously monitor heart rate, vascular activity, and breathing rhythm without physical contact [23].
• Local Data Processing: On the edge device, process raw biosignals locally to extract features for cycle phase prediction. Raw data never leaves the device [23].
• Federated Model Training:
  • Initialize a global machine learning model (e.g., deep neural network) on the central server.
  • Edge devices download the global model and improve it using local data.
  • Only model parameter updates (not raw data) are sent back to the server for aggregation.
  • The server averages updates to create an improved global model, iterating this process [23].
• Personalized Prediction: The framework dynamically adapts to individual physiological patterns and irregular cycles, providing real-time phase predictions on the local device [23].

Protocol for Phase Determination with Hormonal Confirmation

Objective: To accurately determine menstrual cycle phases using a combination of self-report and hormonal assays, minimizing misclassification [10] [6].

Materials:

• Materials for hormone assessment (saliva, blood serum/plasma, or urine kits) [10] [6].
• Standardized daily symptom tracking logs or electronic diaries.
• Equipment for hormone assay (e.g., ELISA kits, mass spectrometry) [6].

Procedure:

• Forward Calculation for Visit Scheduling: For initial phase estimation, use forward counting from the first day of menstruation (Cycle Day 1) based on a typical 28-day cycle or participant's reported average cycle length [10] [6].
• Hormone Sampling for Phase Confirmation: Collect hormone samples at multiple time points across the cycle, ideally at least 3 observations per person to model within-person variance [6].
• Phase Definition via Hormone Levels: Define phases based on established hormone level thresholds and ratios, rather than self-report alone [10] [6]:
  • Late Follicular/Ovulatory Phase: Characterized by rising or peak estradiol levels and low progesterone.
  • Mid-Luteal Phase: Characterized by sustained high progesterone with a secondary estradiol peak.
  • Menstrual Phase: Characterized by low levels of both estradiol and progesterone.
• Data Exclusion: Exclude anovulatory cycles (confirmed by lack of progesterone rise) from analysis [6].

Visualization of Workflows and Signaling Pathways

Data Source Integration Workflow

Data Source Integration Workflow

Physiological Signaling Pathways in Cycle Tracking

Physiological Signaling Pathways

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials for Menstrual Cycle Phase Determination Research

Item	Function/Application	Examples/Specifications
Wearable Sensors	Continuous physiological data collection under free-living conditions [20] [21]	Wrist-worn (Ava Bracelet, Oura Ring, Empatica EmbracePlus); Intravaginal (OvulaRing) [20] [25] [11]
Urinary LH Tests	Ground truth confirmation of ovulation and fertile window closure [21] [11]	Clearblue Digital Ovulation Test; Used daily from cycle day 6 until LH surge detection [21] [11]
Hormone Assay Kits	Quantification of estradiol and progesterone levels for phase confirmation [10] [6]	Saliva, blood serum/plasma, or urine-based ELISA kits; Mass spectrometry for high precision [10] [6]
Electronic Diaries/EMA Platforms	Tracking confounders, symptoms, menstrual start dates, and participant compliance [21] [6]	Custom smartphone apps; Ecological Momentary Assessment (EMA) platforms; Must capture activity, stress, sleep, diet [21] [6]
Federated Learning Framework	Privacy-preserving model training across decentralized data sources [23]	Custom software integrating edge AI with secure model aggregation; Enables analysis of sensitive health data without centralization [23]
Data Processing & ML Software	Analysis of high-dimensional physiological time-series data for phase classification [18] [11]	Python/R with scikit-learn, XGBoost, TensorFlow/PyTorch; Specialized for time-series analysis and multilevel modeling [18] [11]

Quantitative Data Synthesis

The following tables summarize quantitative changes in key physiological inputs across the menstrual cycle phases, as reported in the literature. These values are essential for establishing phase-specific baselines in research protocols.

Table 1: Heart Rate Variability (HRV) and Cardiac Parameters Across Menstrual Cycle Phases

Parameter	Menstrual Phase (Mean ± SD)	Proliferative/Follicular Phase (Mean ± SD)	Secretory/Luteal Phase (Mean ± SD)	Statistical Significance & Notes
Mean Heart Rate (bpm)	79.08 ± 8.84	73.87 ± 8.96	Higher than other phases	Significant difference between Menstrual-Proliferative and Proliferative-Secretory [26]
Mean RR Interval (ms)	0.75 ± 0.13	0.18 ± 0.15	Higher in secretory phase	Significant in Proliferative-Secretory and Secretory-Menstrual comparisons [26]
RMSSD (ms)	27.78 ± 19.84	31.87 ± 21.22	Lower in secretory phase	Statistically significant change between Proliferative and Secretory phases [26]
LF Power (nu)	72.41 ± 9.19	65.93 ± 13.93	Higher in secretory phase	Significant in all phase comparisons; indicates sympathetic activity [26]
HF Power (nu)	27.59 ± 9.19	34.06 ± 13.93	Lower in secretory phase	Significant in all phase comparisons; indicates parasympathetic activity [26]
LF/HF Ratio	3.01 ± 1.35	2.47 ± 1.49	Higher in secretory phase	Suggests sympathetic predominance in secretory/luteal phase [26]

Table 2: Skin Temperature and Other Physiological Parameters Across Menstrual Cycle Phases

Parameter	Follicular Phase	Luteal Phase	Statistical Significance & Notes
Nocturnal Finger Skin Temp (°C)	Baseline	+0.30 ± 0.12	Significantly higher in luteal phase (p<0.001) [27]
Oral BBT (°C)	Baseline	+0.23 ± 0.09	Significantly higher in luteal phase (p<0.001) [27]
Maximal Breath-Hold Time (s)	106.10 ± 12.42	115.59 ± 13.95	Significantly higher in mid-luteal phase (p<0.001) [28]
Metabolic Rate & CO2 Buildup	Higher	Lower	Significantly higher in early follicular phase (p<0.001) [28]

Experimental Protocols

Protocol for Heart Rate Variability (HRV) Assessment

This protocol is adapted from studies investigating autonomic tone across the menstrual cycle [26] [29].

Aim: To measure cardiac autonomic function via HRV in different phases of the menstrual cycle. Design: Cross-sectional observational study with repeated measures.

Methodology:

• Participants: Recruit healthy, eumenorrheic women (e.g., aged 18-25). Exclude those with endocrine disorders, pregnancy, irregular cycles, or using hormonal contraception.
• Phase Determination: Schedule testing sessions for three key phases based on a menstrual diary and/or hormonal assays:
  • Menstrual Phase: Day 2 of cycle (onset of menses).
  • Proliferative/Follicular Phase: ~Day 10.
  • Secretory/Luteal Phase: ~Day 21.
• Prerequisites:
  • Conduct measurements at the same time of day for each participant to control for diurnal variation.
  • Ensure participants are in a rested, supine position in a quiet environment.
  • Avoid caffeine, strenuous exercise, and alcohol for at least 12 hours prior to testing.
• Data Acquisition:
  • Record ECG for 5 minutes using a standard setup (e.g., Physio Pac/PC-2004) or an HR monitor (e.g., Polar H10).
  • Extract RR intervals (beat-to-beat) from the recording.
• HRV Analysis:
  • Time Domain Analysis: Calculate SDNN, RMSSD, NN50, and pNN50.
  • Frequency Domain Analysis: Use Fast-Fourier Transformation to compute Very Low Frequency (VLF), Low Frequency (LF), and High Frequency (HF) power in absolute and normalized units (nu). Calculate the LF/HF ratio.
• Statistical Analysis: Use paired t-tests to compare parameters across phases, with statistical significance set at p<0.05.

Protocol for Continuous Skin Temperature Monitoring

This protocol outlines the use of wearable devices for tracking menstrual cycle rhythms via skin temperature [30] [27].

Aim: To utilize nocturnal finger skin temperature for menstrual cycle phase tracking and predicting ovulation and menstruation. Design: Longitudinal, ambulatory pilot study.

Methodology:

• Participants: Recruit premenopausal, ovulating women not using hormonal contraception.
• Device: Use a wearable temperature sensor (e.g., Oura Ring) worn on the finger.
• Data Collection:
  • Participants wear the device continuously, especially during sleep.
  • Data is automatically synced via Bluetooth to a companion application each morning.
• Reference Measures:
  • Urinary Ovulation Tests: Participants self-test for Luteinizing Hormone (LH) surge to confirm ovulation.
  • Menstrual Diary: Participants record start and end days of menstruation.
  • Oral BBT (optional validation): Measure oral temperature immediately upon waking before any activity.
• Data Processing:
  • Process minute-by-minute nocturnal temperature data.
  • Apply a moving average filter (e.g., 17-min) to raw data from a defined nocturnal window (e.g., 10:00 PM to 8:00 AM).
  • Extract a single, stable, representative nightly temperature value (e.g., the highest stable filtered temperature).
• Cycle Phase Detection & Prediction:
  • Apply a cosine model (Cosinor) or biphasic square wave model to the time series of nightly temperatures to identify the temperature shift indicating ovulation [30].
  • Develop algorithms to predict the start of menstruation (sensitivity of 71.9–86.5% within a ±2-4 day window) and ovulation (sensitivity up to 83.3% within a fertile window) [27].

Protocol for Multi-Parameter Phase Classification using Machine Learning

This protocol leverages multiple physiological signals from wearables with machine learning for automated phase identification [11].

Aim: To classify menstrual cycle phases using physiological signals from a wrist-worn device. Design: Observational study collecting longitudinal data from multiple cycles.

Methodology:

• Participants & Device: Recruit participants to wear a multi-sensor wristband (e.g., measuring skin temperature, electrodermal activity, interbeat interval, heart rate, accelerometry).
• Ground Truth Labeling: Define cycle phases using a combination of:
  • LH Surge: Urinary LH tests to pinpoint ovulation.
  • Menstrual Diary: Self-reported start of menses.
  • Phases: Define as Menses (Period), Follicular (pre-LH surge), Ovulation (e.g., -2 to +3 days around LH surge), Luteal (post-ovulation) [11].
• Feature Extraction:
  • Fixed Window: Extract features (e.g., mean, standard deviation) from non-overlapping windows for each cycle phase.
  • Rolling Window: Extract features using a sliding window for daily phase tracking.
• Model Training & Evaluation:
  • Use algorithms like Random Forest (RF) for classification.
  • Employ validation methods such as "leave-last-cycle-out" or "leave-one-subject-out."
  • Expected Performance: RF models can achieve ~87% accuracy for 3-phase (P, O, L) classification and ~68% accuracy for daily 4-phase (P, F, O, L) tracking [11].

Signaling Pathways and Workflow Visualizations

Menstrual Cycle Hormonal Signaling Pathway

Experimental Workflow for Physiological Data Collection

Autonomic Nervous System Balance Shift

Research Reagent Solutions

Table 3: Essential Materials and Reagents for Menstrual Cycle Physiology Research

Item	Function/Application	Example Products/Assays
Wearable Physiological Monitor	Continuous, ambulatory measurement of HR, HRV, skin temperature, and activity.	Oura Ring, Fitbit Sense, Ava Bracelet, Empatica E4 [27] [11] [31]
Electrocardiogram (ECG) Monitor	High-fidelity recording of heart signals for precise HRV analysis in lab settings.	Polar H10 chest strap, Physio Pac/PC-2004 with ECG electrodes [26] [29]
Urinary Luteinizing Hormone (LH) Test	Reference method for pinpointing the LH surge and confirming ovulation.	At-home LH test kits (e.g., Mira Plus) [27] [11]
Hormone Analyzer	Quantitative measurement of estrogen (E3G) and progesterone (PdG) metabolites in urine.	Mira Plus Starter Kit [32]
Data Analysis & ML Software	For statistical analysis of physiological data and training classification models.	R, Python (with scikit-learn), MATLAB [27] [11]
Basal Body Temperature (BBT) Thermometer	Traditional method for tracking biphasic temperature shift; used for validation.	Digital oral thermometer (e.g., Omron Ecotemp Basic) [27]

The application of machine learning (ML) in biomedical research is transforming how we analyze complex physiological data, particularly in areas like women's health where multifactorial parameters interact. This document provides application notes and experimental protocols for three prominent ML classifiers—Random Forest, XGBoost, and Neural Networks—within the context of menstrual cycle phase classification research. This domain presents unique challenges including high-dimensional temporal data, individual variability, and the need for non-invasive measurement techniques, making it an ideal testbed for comparing classifier efficacy. The protocols outlined below are designed for researchers, scientists, and drug development professionals working to develop precise, personalized health monitoring solutions.

Table 1: Performance Comparison of ML Classifiers in Menstrual Cycle Research

Classifier	Application Context	Reported Accuracy	Key Strengths	Citation
Random Forest (RF)	4-phase classification (P,F,O,L) using wristband data	71% (4-phase); 87% (3-phase)	Robust to overfitting, handles mixed data types	[11]
XGBoost	Ovulation detection using sleeping heart rate	Significant improvement over baseline	Handles temporal dependencies, robust to sleep timing variability	[33]
Random Forest	Ovulation prediction from physiological features	74% (intraday); near-perfect (interday)	Identifies most predictive feature subsets	[34]
Support Vector Machine (SVM)	PCOS diagnosis using pulse wave & TCM indices	83.7% accuracy, AUC=0.878	Effective for clinical index integration	[35]

Classifier Fundamentals and Applications

Random Forest

Random Forest operates as an ensemble method that constructs multiple decision trees during training and outputs the mode of classes (classification) or mean prediction (regression) of individual trees. Its inherent resistance to overfitting makes it particularly suitable for biomedical datasets where features often exceed sample sizes.

In menstrual cycle research, Random Forest has demonstrated exceptional performance. One study achieving 87% accuracy in three-phase classification (menstruation, ovulation, luteal) used physiological signals from wrist-worn devices including skin temperature, electrodermal activity, interbeat interval, and heart rate [11]. The algorithm's capability to rank feature importance provides additional scientific value, revealing which physiological parameters most strongly predict cycle phases.

XGBoost (Extreme Gradient Boosting)

XGBoost represents an advanced implementation of gradient boosting that sequentially builds decision trees, where each tree corrects errors of its predecessor. Its computational efficiency, handling of missing values, and regularization to prevent overfitting make it ideal for processing continuous wearable sensor data.

Recent research has demonstrated XGBoost's particular strength in detecting subtle circadian rhythm patterns in heart rate data for ovulation detection. The model successfully identified the heart rate at circadian rhythm nadir (minHR) as a novel feature that significantly improved luteal phase classification, especially in individuals with high variability in sleep timing where it outperformed traditional basal body temperature methods by reducing absolute errors in ovulation detection by 2 days [33].

Neural Networks

While the search results provided limited specific performance metrics for neural networks in menstrual cycle classification, their capacity to model complex non-linear relationships in high-dimensional data makes them theoretically suitable for this domain. Earlier research has demonstrated their application in related women's health contexts, such as using deep residual neural networks (ResNet) for classifying menstrual phases based on pulse signal data, achieving 81.8% accuracy in personalized models [11].

Experimental Protocols for Menstrual Cycle Phase Classification

Data Acquisition and Preprocessing Protocol

Table 2: Essential Research Reagent Solutions

Component	Specification	Research Function
Wearable Sensors	Empatica E4, EmbracePlus, Oura Ring	Collects physiological signals (HR, HRV, temperature, EDA) in free-living conditions
Urinary LH Tests	Commercial immunoassay kits	Provides ground truth for ovulation timing
Data Processing Pipeline	Python/R with signal processing libraries	Filters artifacts, extracts features from raw sensor data
Feature Selection Algorithm	Recursive Feature Elimination with Cross-Validation (RFECV)	Identifies most predictive parameters from high-dimensional data

Procedure:

• Participant Recruitment: Enroll healthy premenopausal women (18-35 years) with regular menstrual cycles (21-35 days). Exclude participants using hormonal contraception or with conditions affecting menstrual cyclicity.
• Sensor Deployment: Distribute FDA-cleared wearable devices (Empatica E4, Oura Ring, or equivalent) configured to collect data at appropriate frequencies (e.g., heart rate every 5 minutes, skin temperature continuously).
• Ground Truth Establishment: Implement urinary luteinizing hormone (LH) testing protocol with daily testing beginning cycle day 10 until positive test confirms ovulation.
• Data Collection: Acquire continuous physiological data over multiple complete menstrual cycles (minimum 2-3 cycles per participant) to capture intra-individual variability.
• Signal Preprocessing: Apply bandpass filters to remove artifacts, detect and interpolate missing values using established algorithms, and normalize data per participant baselines.
• Feature Extraction: Calculate domain-specific features including:
  • Heart rate variability metrics (SDNN, RMSSD)
  • Circadian rhythm parameters (acrophase, amplitude, minHR)
  • Skin temperature trends and nocturnal declines
  • Sleep architecture metrics (when available)

Model Training and Validation Protocol

Procedure:

• Data Labeling: Annotate cycles according to reference definitions:
  • Menstrual phase: Days of active bleeding
  • Follicular phase: Post-menstruation to pre-ovulation
  • Ovulatory phase: ±3 days around positive LH test
  • Luteal phase: Post-ovulation to next menses
• Data Partitioning: Implement leave-last-cycle-out or leave-one-subject-out cross-validation to rigorously assess generalizability and prevent data leakage.
• Feature Selection: Apply recursive feature elimination with cross-validation (RFECV) to identify the most predictive parameters from the high-dimensional feature set.
• Model Training:
  • For Random Forest: Optimize number of trees (100-500), maximum depth, and minimum samples per leaf
  • For XGBoost: Tune learning rate (0.01-0.3), maximum depth (3-10), and subsampling ratios
  • For Neural Networks: Architect appropriate network topology based on data structure (CNN for temporal patterns, LSTM for sequential dependencies)
• Model Validation: Assess performance using accuracy, precision, recall, F1-score, and area under receiver operating characteristic curve (AUC-ROC) with strict separation between training, validation, and test sets.

Implementation Workflows

Figure 1: End-to-End Workflow for Menstrual Cycle Classification Development

Figure 2: Comparative Architecture of Random Forest vs. XGBoost Approaches

The structured comparison and protocols provided herein establish a foundation for implementing machine learning classifiers in menstrual cycle research. Random Forest and XGBoost have demonstrated particular efficacy in this domain, offering robust performance while providing interpretability through feature importance metrics. These approaches enable non-invasive, continuous cycle phase monitoring that surpasses traditional methods in accuracy and practicality, especially for individuals with irregular sleep patterns or cycle variability.

For drug development professionals, these methodologies offer potential applications in clinical trial participant screening, monitoring treatment effects on menstrual cyclicity, and personalizing therapeutic interventions based on cycle phase. The integration of these classifiers into clinical decision support systems, as demonstrated in reproductive medicine [36], highlights their translational potential in women's health innovation.

Within the burgeoning field of women's health research, particularly in the coding of menstrual cycle day phase protocols, the analysis of temporal physiological data is paramount. The advent of wearable technology has enabled the continuous collection of high-frequency data streams, such as heart rate (HR), skin temperature, and electrodermal activity (EDA). Transforming these raw, longitudinal datasets into meaningful inputs for predictive models requires sophisticated feature engineering techniques. The choice between fixed windows and rolling windows for temporal aggregation is a critical methodological decision that directly impacts the accuracy, generalizability, and clinical applicability of phase classification models. This document outlines formal application notes and protocols for employing these techniques in menstrual cycle research.

Definitions and Core Concepts

Fixed Time Windows

A fixed window (or calendar window) approach segments data into non-overlapping, consecutive blocks of time defined a priori based on the underlying biological process [37]. In menstrual cycle research, this often means segmenting data into phases—such as menstrual, follicular, ovulatory, and luteal—based on a reference point like a positive luteinizing hormone (LH) test or the first day of menses [11]. Features (e.g., mean, max, standard deviation) are then calculated independently for each of these static segments.

Rolling Time Windows

A rolling window (or moving window) technique computes metrics over a fixed-size window of observations that "rolls" or slides sequentially through the time series data [38]. With each new time step, the window discards the oldest data point and incorporates the newest one [39] [40]. This method is used to create a series of localized measurements, such as a 7-day rolling average of nightly skin temperature, which smooths out short-term noise and can highlight underlying physiological trends [38] [41].

Comparative Analysis in Menstrual Cycle Research

The selection of a windowing technique involves trade-offs between noise reduction, temporal precision, and biological validity. The following table summarizes a quantitative comparison derived from a study that explicitly tested both methods for classifying menstrual cycle phases using physiological data from a wrist-worn device [11].

Table 1: Performance Comparison of Fixed vs. Rolling Windows for Menstrual Phase Classification

Aspect	Fixed Window Technique	Rolling Window Technique
General Approach	Segments data into pre-defined, non-overlapping physiological phases (e.g., P, F, O, L) [11].	Uses a sliding window that moves through the time series, often day-by-day [11].
Reported Performance (3-phase model)	87% accuracy, AUC-ROC: 0.96 [11]	Information not available in provided search results.
Reported Performance (4-phase model)	71% accuracy, AUC-ROC: 0.89 [11]	68% accuracy, AUC-ROC: 0.77 [11]
Temporal Alignment	Aligns with clinical/biological event markers (e.g., LH surge).	Agnostic to underlying phase boundaries; provides a continuous output.
Primary Advantage	Directly models the biphasic or triphasic nature of the cycle, often yielding higher accuracy for phase classification [11].	Higher resolution tracking; can potentially capture transitions between phases more smoothly.
Primary Disadvantage	Requires accurate ground-truth labeling for each phase, which can be burdensome (e.g., LH tests) [42].	May be noisier and less accurate for definitive phase classification [11].

Experimental Protocols

Protocol for Fixed Window Feature Engineering

This protocol is ideal for hypothesis testing where the phase boundaries are known from reference methods.

• Data Labeling: Define cycle phases based on a reference standard. For example [11]:
  • Menses (P): Days with recorded menstrual bleeding.
  • Ovulation (O): The period spanning 2 days before to 3 days after a positive LH test.
  • Luteal (L): The phase following ovulation until the start of the next menses.
• Segmentation: Partition the entire time series of each physiological signal (temperature, HR, etc.) according to the defined phase labels.
• Feature Extraction: For each signal within each fixed window, calculate summary statistics. Example features include:
  • Mean and standard deviation of nocturnal skin temperature.
  • Minimum heart rate during sleep.
  • Mean of the tonic component of Electrodermal Activity (EDA).
• Model Training: Use the extracted features (e.g., one row per cycle phase) to train a classifier, such as a Random Forest, for phase identification [11].

Protocol for Rolling Window Feature Engineering

This protocol is suitable for developing daily prediction models or for applications where phase boundaries are not known a priori.

• Window Definition: Select a window size (e.g., 5 days, 7 days). The size should be large enough to smooth noise but small enough to capture relevant physiological shifts [39] [38].
• Feature Extraction: For each day in the time series, calculate statistics for the preceding n days (the window). For a 5-day window rolling daily, you would calculate [41] [43]:
  • Rolling mean of skin temperature.
  • Rolling standard deviation of heart rate.
  • Rolling minimum of the inter-beat interval (IBI).
• Data Structuring: Each day becomes a data point with features representing the rolling statistics of the preceding window.
• Model Training & Prediction: Train a model (e.g., Logistic Regression) to predict a phase label for each day based on its rolling window features. This generates a daily, dynamic prediction of cycle phase [11].

Visualization of Workflows

The following diagram illustrates the logical workflow for choosing and applying these techniques within a menstrual cycle research study.

The Scientist's Toolkit: Research Reagent Solutions

The following table details key materials, software, and analytical methods essential for implementing the described feature engineering protocols.

Table 2: Essential Reagents and Tools for Temporal Analysis of Menstrual Cycle Data

Item Name	Type	Function/Application in Research
Wrist-worn Wearable Device (e.g., E4, EmbracePlus)	Hardware	Enables continuous, passive collection of physiological signals including skin temperature, heart rate (HR), inter-beat interval (IBI), and electrodermal activity (EDA) [11].
Luteinizing Hormone (LH) Urinary Test Kits	Biochemical Assay	Provides the reference standard for pinpointing the day of ovulation, which is critical for creating accurate labels for fixed window segmentation [11] [42].
Python Pandas Library	Software Library	The primary tool for data manipulation, including the implementation of rolling window calculations (.rolling()) and the creation of lagged features (.shift()) [44] [41].
Circular Statistics	Analytical Method	A branch of statistics for analyzing periodic data (e.g., a ~28-day cycle). Used to test for significant periodicity in physiological features across the menstrual cycle [42].
Random Forest Classifier	Machine Learning Model	A powerful ensemble algorithm frequently used for phase classification tasks; demonstrated high performance with fixed-window features in menstrual cycle studies [11].
Autoregressive Integrated Moving Average (ARIMA)	Statistical Model	Used for time series forecasting of physiological signals (e.g., predicting next day's temperature), which can itself be a feature for phase classification [42].

Within menstrual cycle research, a fundamental tension exists between developing generalized, population-level models and creating personalized, individualized algorithms. The choice between these strategies directly impacts the accuracy, reliability, and clinical applicability of research findings. This protocol examines two predominant methodological frameworks: Leave-One-Subject-Out (LOSO) cross-validation and population-level modeling, providing researchers with standardized approaches for implementing each strategy within menstrual cycle studies. The physiological complexity of the menstrual cycle—characterized by significant inter-individual variability in hormone responses, cycle length, and symptomatology—necessitates rigorous methodological standards to ensure valid and replicable findings [6]. By establishing clear protocols for both personalized and population-level approaches, this document aims to enhance methodological consistency across the field and facilitate more meaningful cross-study comparisons.

Comparative Performance Analysis

Table 1: Quantitative Performance Comparison of Modeling Approaches

Modeling Approach	Cycle Phase Classification	Accuracy (%)	AUC-ROC	Use Case Context
Population-Level (Leave-Last-Cycle-Out)	3 Phases (P, O, L)	87	0.96	Initial model development, general pattern identification [11]
Population-Level (Leave-Last-Cycle-Out)	4 Phases (P, F, O, L)	71	0.89	Fine-grained phase differentiation [11]
Leave-One-Subject-Out (LOSO)	3 Phases (P, O, L)	87	N/R	Assessment of model generalizability across new individuals [11]
Leave-One-Subject-Out (LOSO)	4 Phases (P, F, O, L)	63	N/R	Testing robustness to individual physiological variability [11]

N/R = Not Reported in the source material.

The performance differential between population-level and LOSO approaches reveals a critical trade-off between overall accuracy and generalizability. Population-level models, trained on aggregated data from multiple participants, demonstrate superior performance when tested on data from the same population, achieving up to 87% accuracy for three-phase classification [11]. However, this approach risks overfitting to population-specific characteristics and may not adequately capture the substantial physiological variability between individuals. In contrast, the LOSO approach, which iteratively trains models on all but one subject and tests on the held-out subject, provides a more rigorous assessment of model generalizability across new individuals. While LOSO typically yields lower absolute accuracy metrics (63% for four-phase classification), it more accurately represents real-world performance where models encounter entirely new subjects with unique physiological signatures [11].

Methodological Protocols

Participant Screening and Data Collection Protocol

• Inclusion Criteria: Establish clear criteria for participant selection, including age range (typically 18-45), regular menstrual cycles (21-35 days), absence of hormonal contraceptive use, and no known reproductive disorders [6]. Document all exclusion criteria transparently.
• Cycle Confirmation: Utilize ovulation confirmation tests (e.g., urinary luteinizing hormone kits) to objectively identify ovulation and define cycle phases, rather than relying solely on calendar-based estimates [11] [6].
• Data Collection Schedule: Implement frequent sampling protocols (daily or multi-daily) to capture dynamic physiological changes. For difficult-to-collect data types (e.g., psychophysiological measures), strategically schedule assessments across key cycle phases [6].
• Wearable Device Configuration: Deploy FDA-cleared or research-grade wearable devices (e.g., Empatica E4, Oura ring) to collect continuous physiological data including heart rate (HR), interbeat interval (IBI), heart rate variability (HRV), skin temperature, and electrodermal activity (EDA) [11] [45] [33].

Data Preprocessing and Feature Engineering

• Signal Processing: Apply Butterworth low-pass filters to raw physiological signals to remove high-frequency noise while preserving biologically relevant information [45]. Implement linear interpolation for minimal missing data points (<5%).
• Feature Extraction: Calculate both standard and novel features from cleaned signals. For heart rate data, derive the heart rate at the circadian rhythm nadir (minHR), which demonstrates particular utility for luteal phase classification and ovulation detection, especially in individuals with variable sleep patterns [33].
• Cycle Alignment: Normalize cycle length to a standard duration (e.g., 28 days) to account for inter-individual variability in follicular phase length while preserving the more consistent luteal phase duration [45].
• Phase Definition: Adopt consistent, biologically-grounded definitions for menstrual cycle phases based on ovulation confirmation rather than calendar estimates alone [6].

Model Training and Validation Implementation

• Algorithm Selection: Implement multiple algorithm classes including Random Forest, XGBoost, and logistic regression to compare performance across different model architectures [11] [33].
• LOSO Implementation: For LOSO validation, iteratively train models on data from all but one participant and test on the completely held-out participant. Repeat this process until each participant has served as the test set once [11].
• Population-Level Validation: For population-level approaches, implement leave-last-cycle-out validation where models are trained on initial cycles and tested on the final cycle from each participant, or use standard train-test splits with appropriate stratification [11].
• Performance Metrics: Report multiple performance metrics including accuracy, precision, recall, F1-score, and AUC-ROC to provide a comprehensive assessment of model performance across different aspects [11].

The Scientist's Toolkit

Table 2: Essential Research Reagents and Materials

Category	Specific Tools/Reagents	Research Application
Wearable Sensors	Empatica E4, Oura Ring, Huawei Band 5	Continuous physiological monitoring (HR, IBI, HRV, temperature, EDA) in free-living conditions [11] [46]
Ovulation Confirmation	Urinary LH Test Kits	Objective identification of ovulation timing for phase definition and model validation [11] [6]
Data Processing	Python, R, MATLAB	Signal processing, feature extraction, and machine learning implementation [11] [45]
Algorithm Libraries	Scikit-learn, XGBoost	Implementation of Random Forest, XGBoost, and other ML algorithms for classification [11] [33]
Statistical Analysis	R, SAS, SPSS	Multilevel modeling to account for nested data structure (cycles within individuals) [6]

Experimental Implementation Workflow

Discussion and Strategic Recommendations

The comparative analysis between LOSO and population-level approaches reveals distinct advantages and limitations for each method. Population-level models excel in contexts where general patterns are sufficient and computational efficiency is prioritized, achieving up to 87% accuracy for three-phase classification [11]. However, these models demonstrate reduced performance (63% accuracy in LOSO validation) when applied to new individuals, highlighting their limited generalizability across diverse populations [11]. Conversely, LOSO approaches provide a more robust assessment of model performance in real-world scenarios where algorithms must generalize to entirely new individuals, making them particularly valuable for clinical applications and personalized health monitoring.

Based on empirical findings, we recommend:

• Population-level models for initial exploratory analysis, general pattern identification, and contexts where the target population closely matches the training cohort.
• LOSO validation for assessing real-world generalizability, clinical application development, and research involving heterogeneous populations.
• Hybrid approaches that combine population-level training with personalized fine-tuning, potentially leveraging transfer learning techniques to adapt general models to individual physiological patterns [11].

Future methodological development should focus on refining personalized modeling approaches, particularly through transfer learning and adaptive algorithms that can continuously refine predictions based on individual data streams. Additionally, standardized reporting of validation methodologies will enhance cross-study comparisons and accelerate methodological advances in menstrual cycle research.

Troubleshooting and Optimization: Enhancing Model Performance and Robustness

Addressing the Challenge of Irregular Cycles and Anovulation

Irregular menstrual cycles and anovulation present significant methodological challenges in endocrine and drug development research. These conditions, clinically categorized under Abnormal Uterine Bleeding associated with Ovulatory Dysfunction (AUB-O), introduce substantial variability that complicates the standardization of cycle phase protocols essential for rigorous scientific investigation [47]. The hypothalamic-pituitary-ovarian (HPO) axis disruption underlying these conditions leads to unpredictable hormone profiles and cycle lengths, rendering standard "count-forward" or "count-backward" phase estimation methods highly unreliable [10].

Accurately identifying and classifying these cycles is paramount for research integrity. It ensures appropriate participant stratification, enables the detection of meaningful biobehavioral correlates of ovarian hormones, and is crucial for clinical trials where cycle phase may affect drug metabolism or therapeutic outcomes [6] [12]. This document provides detailed application notes and standardized protocols to address these challenges, facilitating more reproducible and valid research outcomes.

Clinical Definitions and Etiologies

Defining Irregularities and Anovulation

• Abnormal Uterine Bleeding-Ovulatory Dysfunction (AUB-O): A condition characterized by irregular, prolonged, and often heavy menstrual bleeding due to the failure of ovulation. Without ovulation, progesterone is not produced, resulting in unopposed estrogen stimulation of the endometrial lining [47].
• Anovulatory Bleeding: Irregular, non-cyclic uterine bleeding occurring in the absence of ovulation. It is often interspersed with phases of amenorrhea [47].
• Cycle Length Variability: Regular cycles typically range from 21 to 38 days. Irregular cycles are consistently shorter than 24 days (polymenorrhea) or longer than 38 days (oligomenorrhea), with high cycle-to-cycle variability [6] [13].
• Ovulatory Confirmation: The gold standard for confirming ovulation requires a combination of methods, including a detectable urinary luteinizing hormone (LH) surge, a sustained rise in basal body temperature (BBT), a mid-luteal phase progesterone elevation (in serum or urine), and serial ultrasonography showing follicular collapse [13] [11].

Common Etiologies of AUB-O

The following table summarizes primary etiologies and their research considerations.

Table 1: Common Etiologies of Anovulation and Irregular Cycles

Etiology Category	Specific Conditions / Examples	Key Research Considerations
Physiological	Perimenarche, Perimenopause, Lactation	Expected life-stage transitions; requires distinct stratification [47].
Endocrine Disorders	Polycystic Ovary Syndrome (PCOS), Thyroid Dysfunction, Hyperprolactinemia	Common causes of pathological anovulation; PCOS is a major focus [47] [13].
Energy Balance & Lifestyle	Anorexia, Excessive Exercise (e.g., Athletes), Relative Energy Deficiency in Sport (RED-S)	Low body mass index (BMI) or high exercise load is a key risk factor; common in athlete populations [47] [13].
Medications	Antiepileptics (e.g., Valproate), Antipsychotics (e.g., Haloperidol, Risperidone)	Iatrogenic cause; must be carefully screened for and documented [47].
Psychological Stress	Chronic high stress	Can disrupt HPO axis function [47].

Standardized Assessment and Classification Protocol

A multi-modal approach is critical for accurately classifying cycle status in research participants. Reliance on self-reported cycle history alone is insufficient.

Algorithm for Participant Assessment

The following diagram outlines a logical workflow for assessing and classifying participants in a research study.

Detailed Methodologies for Key Experiments

Protocol 1: Prospective Cycle Tracking and Symptom Monitoring

• Purpose: To establish a reliable baseline of cycle length variability and symptom patterns.
• Materials: Dedicated paper diary or secure digital platform (e.g., custom app); basal body thermometer (digital, BBT-specific).
• Procedure:
  • Participants log daily bleeding/spotting intensity for a minimum of two full cycles [6].
  • Participants measure and log BBT immediately upon waking, before any physical activity [11].
  • For mood and behavioral studies, participants complete daily ratings of specific symptoms (e.g., using the Carolina Premenstrual Assessment Scoring System (C-PASS) for PMDD/PME diagnosis [6]).
• Data Analysis: Calculate cycle length variability (standard deviation). Visually inspect BBT charts for a biphasic pattern indicative of ovulation. Analyze daily symptom ratings to identify cyclical patterns.

Protocol 2: Urinary Hormone Monitoring for Ovulation Confirmation

• Purpose: To objectively predict and confirm the occurrence and timing of ovulation.
• Materials: Quantitative urine hormone monitor (e.g., Mira monitor, Clearblue Monitor) and corresponding test strips for LH and PdG [16] [13].
• Procedure:
  • Participants begin daily testing from cycle day 6 until ovulation is confirmed.
  • A surge in LH levels is used to predict imminent ovulation (within ~24-48 hours).
  • A sustained rise in PdG (pregnanediol glucuronide, a urinary metabolite of progesterone) > 5 μg/mL for several days following the LH surge is used to confirm that ovulation has occurred [13].
• Data Analysis: The LH peak is defined as the highest value preceding a sustained rise in PdG. The day of the LH peak is typically designated as "ovulation day."

Protocol 3: Gold Standard Ultrasound Validation Protocol

• Purpose: To provide the definitive assessment of follicular development and ovulation for validating other methods.
• Materials: Ultrasound machine with transvaginal transducer.
• Procedure:
  • Participants undergo serial ultrasounds, beginning around cycle day 10-12 or when a leading follicle reaches ~14mm [13].
  • Scans are repeated every 1-2 days until follicular rupture (ovulation) is observed, characterized by the sudden disappearance or reduction in size of the dominant follicle, often with fluid in the cul-de-sac.
• Data Analysis: The day of ovulation is defined as the day following the last identification of the dominant follicle [13]. This data serves as the reference point for evaluating the accuracy of urinary hormone predictions.

Quantitative Data and Validation

Performance of Common Phase Determination Methods

Common methods for determining menstrual cycle phase are highly error-prone when applied to individuals, especially in the context of irregular cycles. The following table summarizes quantitative findings on their accuracy.

Table 2: Accuracy of Common Menstrual Cycle Phase Determination Methods

Method Category	Specific Technique	Reported Performance / Limitations	Source
Self-Report Projection	Forward/Backward Calculation	Error-prone; results in phases being incorrectly determined for many participants. Agreement with hormone-defined phases is low (Cohen’s kappa: -0.13 to 0.53).	[10]
Hormone Ranges	Single time-point serum hormone ranges	Lacks empirical validation; fails to account for individual variability in hormone levels and dynamics. A common but flawed method for "confirming" phase.	[10]
Wearable Sensors & Machine Learning	Random Forest model using wristband data (Temp, HR, EDA, IBI)	3-phase classification (P, O, L): 87% accuracy, AUC 0.96.4-phase classification (P, F, O, L): 71% accuracy, AUC 0.89. Promising for reducing self-report burden.	[11]
Urine Hormone Monitoring	Quantitative tracking of LH and PdG	Correlates well with serum hormone levels and ultrasound day of ovulation. PdG >5 μg/mL used to confirm ovulation. Considered a key tool for at-home monitoring.	[13]

Experimental Workflow for Protocol Validation

The following diagram illustrates the workflow for a validation study comparing quantitative urine hormone monitoring to the ultrasound gold standard.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Tools for Menstrual Cycle Research

Item / Reagent	Function / Purpose	Example Products / Assays
Quantitative Urine Hormone Monitor	At-home tracking of LH and PdG to predict and confirm ovulation. Provides numerical values for pattern analysis.	Mira Fertility Tracker, Inito Fertility Monitor, Clearblue Fertility Monitor [16] [13]
Urine Hormone Test Strips	Disposable strips used with quantitative monitors to assay specific hormones.	Mira Fertility Hormone Test Wands (for FSH, E3G, LH, PdG), Inito test strips [13]
Basal Body Temperature (BBT) Thermometer	Tracking the post-ovulatory biphasic shift in resting body temperature to retrospectively confirm ovulation.	Digital BBT thermometers (e.g., Tempdrop for automated overnight sensing) [16] [11]
LH Urine Ovulation Predictor Kits (Qualitative)	Detecting the LH surge to predict impending ovulation. Provides a qualitative "positive/negative" result.	Clearblue Digital Ovulation Test, Clinical Guard LH Strips [13]
Salivary/Serum Immunoassay Kits	Quantifying estradiol and progesterone levels in saliva or blood serum in a laboratory setting.	Salimetrics ELISA Kits, Roche Diagnostics Electrochemiluminescence Immunoassay (ECLIA) [6] [12]
C-PASS Tool	Standardized system for diagnosing PMDD and Premenstrual Exacerbation (PME) from prospective daily ratings.	Carolina Premenstrual Assessment Scoring System (paper worksheet, Excel, R, or SAS macro) [6]
Validated Daily Symptom Scale	Prospective monitoring of emotional, cognitive, and behavioral symptoms across the cycle.	Daily Record of Severity of Problems (DRSP), custom Visual Analog Scales (VAS) [6]

Mitigating the Impact of Sleep Timing Variability on Signal Quality

In the field of physiological signal research, particularly in longitudinal studies focusing on the menstrual cycle, sleep timing variability represents a significant and often underestimated confounder. This variability—defined as day-to-day fluctuations in sleep onset and wake times—can introduce substantial noise into data, obscuring genuine physiological patterns and compromising the integrity of research findings [48]. The imperative to mitigate its impact is especially critical in drug development and scientific studies where precise phase identification is paramount. This Application Note provides a structured framework, underpinned by quantitative evidence and proven experimental protocols, to enable researchers to identify, quantify, and control for the effects of sleep irregularity, thereby enhancing the signal quality in menstrual cycle and other long-term physiological monitoring studies.

Quantitative Impact of Sleep Variability: Evidence and Data

Emerging research consistently demonstrates that sleep variability is not merely an inconvenience but a key factor with measurable consequences for both physiological signals and psychological outcomes. The data reveal that the stability of sleep patterns can be as influential as total sleep duration.

Table 1: Documented Impacts of Sleep Timing Variability on Key Research Metrics

Metric	Impact of Increased Variability	Quantitative Effect Size	Research Context
Depressive Symptoms	Positive correlation with severity	+0.4 points on PHQ-9 per 1-hr increase in sleep duration SD [48]	Prospective cohort of 2,115 physicians
Next-Day Mood	Negative correlation with improved mood	Negative association with day-to-day shifts in TST and wake time [48]	Same cohort, daily mood assessment
Sleep Quality (Subjective)	Predicts poorer sleep quality	Significant regression weight (b=0.35) for Wake Onset Variability [49]	Healthy adults with 7-9 hour sleep duration
Positive Affect	Predicts reduced positive emotion	Significant regression weight (b=-0.28) for Wake Onset Variability [49]	Same cohort as above
Circadian Phase Stability	Associated with DLMO shift	Correlation (r=0.46) in Delayed Sleep-Wake Phase Disorder [50]	Controlled DLMO assessment study
Wake Time Variability	More variable in DSWPD vs. controls	Significantly higher variability (p≤0.015) [50]	DSWPD patients vs. healthy controls

The posterior cingulate cortex, a key node of the DMN, shows altered functional connectivity with emotion-processing regions like the amygdala and insula in individuals with irregular wake times [49]. This finding provides a neural correlate for the observed emotional and sleep quality deficits associated with sleep variability.

Experimental Protocols for Signal Quality Assurance

To ensure high-fidelity physiological data in the presence of sleep variability, researchers should implement the following standardized protocols.

Protocol for Quantifying and Stratifying Sleep Variability

Objective: To objectively quantify sleep variability and stratify participants or data segments based on irregularity for controlled analysis. Materials: Validated wearable device (e.g., actigraphy watch, Oura Ring, or other research-grade sensor); compatible data analysis software (e.g., MATLAB, R). Procedure:

• Data Collection: Collect continuous sleep-wake data for a minimum of two weeks prior to and throughout the core experimental period. The sampling epoch should be set to 30 or 60 seconds [50].
• Sleep Parameter Extraction: For each 24-hour period, calculate:
  • Sleep Onset Time: Clock time at sleep initiation.
  • Wake Time: Clock time at final awakening.
  • Total Sleep Time (TST): Total duration of sleep within the 24-hour period.
• Variability Calculation: Calculate the intra-individual standard deviation (SD) for each sleep parameter (Sleep Onset, Wake Time, TST) across the data collection window. A useful approach is to use a rolling 7- or 14-day window for dynamic assessment [48].
• Stratification: Classify participants or data epochs into "High Variability" and "Low Variability" groups based on a pre-defined threshold (e.g., median split or Wake Time SD > 1 hour) for subsequent comparative analysis [49] [48].

Protocol for Controlling Sleep Variability in Pre-Phase Longitudinal Studies

Objective: To stabilize circadian phase and minimize sleep-driven signal noise before critical measurements, such as menstrual cycle phase classification. Materials: Wearable sleep tracker; participant instruction sheet; communication tool (e.g., email, app). Procedure:

• Baseline Establishment: Prior to the stabilization phase, have participants maintain their habitual sleep schedule for one week while wearing a sleep tracker to establish individual baseline sleep and wake times.
• Stabilization Phase: Instruct participants to stabilize their sleep schedules for a minimum of four consecutive nights before and during the phase-specific data collection [50].
• Compliance Monitoring: Use actigraphy or other wearable data to verify compliance with the stabilization schedule. Define compliance as sleep onset and wake times within a 60-minute window of their personal target times [50].
• Data Flagging: In the final dataset, flag all physiological measurements (e.g., heart rate, temperature, EEG) taken during periods of high sleep variability (e.g., Wake Time SD > 90 minutes) for sensitivity analysis [48].

Visualization of Workflows

The following diagrams outline the core concepts and methodological workflows discussed in this note.

Impact of Sleep Variability on Research Signals

Mitigation Strategy Workflow

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials and Tools for Sleep Variability Research

Tool / Reagent	Specification / Function	Research Application Example
Research-Grade Actigraph	Worn on the wrist; uses accelerometry to infer sleep and wake states with validated algorithms [50].	Objective longitudinal tracking of sleep onset, wake time, and TST for variability calculation [48].
Prefrontal Portable EEG	2-electrode setup (FP1, FP2); measures neural oscillations like the low α-band (7-8.5 Hz) [51].	Assessing a stable neurobiological correlate of sleep quality (correlated with PSQI) that may be confounded by timing variability [51].
Salivary Melatonin Kits	Materials for sampling saliva in dim light for subsequent assay of melatonin concentration [50].	Determining the Dim Light Melatonin Onset (DLMO), the gold-standard metric for central circadian phase, to assess circadian disruption [50].
Multisensor Wearable (E4, EmbracePlus)	Measures HR, IBI, EDA, and peripheral temperature simultaneously from the wrist [42] [11].	Capturing the physiological signals used for menstrual phase classification while concurrently monitoring sleep-wake patterns.
Validated Sleep/Mood Questionnaires	PSQI for sleep quality; PHQ-9 for depression; Likert scales for daily mood [51] [48].	Quantifying subjective outcomes linked to sleep variability for correlational analysis with objective data.

Data Imputation and Handling Missing Data Points

Within menstrual cycle research, missing data presents a significant challenge that can compromise the validity of scientific findings and drug development outcomes. The menstrual cycle is a dynamic, within-person process characterized by complex hormonal fluctuations, making complete data collection difficult over multiple cycles [6]. This document provides application notes and protocols for handling missing data in menstrual cycle studies, framed within the broader context of coding menstrual cycle day phase protocols for research audiences.

Missing data patterns in menstrual health studies are often non-random. For instance, one study using an Endometriosis Symptom Diary found entries were significantly more likely to be missing on Fridays (18.5%) and Saturdays (22.9%) compared to other days [52]. Understanding these patterns is essential for selecting appropriate imputation methods that maintain the biological integrity of cycle-phase specific analyses.

Menstrual Cycle Characteristics and Data Collection Challenges

Fundamental Cycle Characteristics

Table 1: Real-World Menstrual Cycle Characteristics from Large-Scale Data Analysis

Parameter	Overall Mean	95% Confidence Interval	Variation by Age (25-45 years)	Variation by BMI (>35 vs. 18.5-25)
Cycle Length	29.3 days	Not reported	Decrease of 0.18 days per year	0.4 days (14%) higher variation
Follicular Phase	16.9 days	10-30 days	Decrease of 0.19 days per year	Not specifically reported
Luteal Phase	12.4 days	7-17 days	No significant change	Not specifically reported
Bleed Length	Not reported	Not reported	Decrease of 0.5 days from youngest to oldest	Not specifically reported

Source: Analysis of 612,613 ovulatory cycles from 124,648 users [53]

The substantial natural variation in cycle characteristics highlighted in Table 1 necessitates careful handling of missing data to preserve accurate phase-specific analyses. The luteal phase demonstrates more consistent length (mean: 12.4 days; 95% CI: 7-17 days) compared to the follicular phase, which shows greater variability (mean: 16.9 days; 95% CI: 10-30 days) [53]. This variation is crucial context for imputation decisions.

• Non-compliance with daily protocols: Participants may skip daily entries on weekends or during busy periods [52]
• Inconsistent sleep patterns: Affects basal body temperature (BBT) measurements [18]
• Cycle length variability: Makes fixed-day visit scheduling problematic [54]
• Device-specific issues: Inconsistent wearable sensor data [55]
• Anovulatory cycles: Occurrence of cycles where ovulation does not happen [56]

Methodological Framework for Data Imputation

Foundational Principles

Menstrual cycle research requires specialized consideration for missing data because the cycle is fundamentally a within-person process [6]. Between-subject designs conflate within-subject variance (attributable to changing hormone levels) with between-subject variance (attributable to each individual's baseline symptoms), lacking validity for cycle research [6] [12].

The gold standard approach involves repeated measures designs with daily or multi-daily (ecological momentary assessments) ratings [6]. For statistical modeling of within-person effects, at least three observations per person across one cycle represent the minimal acceptable standard, though three or more observations across two cycles allows for greater confidence in reliability of between-person differences [6].

Data Realignment and Multiple Imputation Protocol

The BioCycle Study established a robust methodology for handling missing menstrual cycle data through realignment and multiple imputation [54]. This approach is particularly valuable when biospecimen collection occurs at carefully timed clinic visits scheduled at key times of hormonal variability.

Experimental Protocol: Data Realignment and Multiple Imputation

Objective: To correctly classify hormonal measurements to biologically relevant menstrual cycle phases and account for missing data generated by the realignment process.

Materials:

• Fertility monitors (e.g., Clearblue Easy Fertility Monitor)
• Hormone assessment kits for estradiol, progesterone, LH, FSH
• Electronic data capture system

Procedure:

• Visit Scheduling: Schedule clinic visits using an algorithm accounting for each participant's self-reported cycle length, with mid-cycle visits adjusted based on fertility monitor data [54].

• Fertility Monitoring: Participants use fertility monitors starting on calendar day 6 after menses, continuing for 10-20 days depending on whether peak levels are detected. Monitors measure estrone-3-glucuronide and LH in urine [54].

• Visit Triggers: If the monitor indicates 'peak fertility' on a day without a scheduled visit, participants come in that morning and the following two mornings [54].

• Hormone Assessment: Collect fasting serum samples at each clinic visit. Measure estradiol by radioimmunoassay, and progesterone, LH, and FSH using solid phase competitive chemiluminescent enzymatic immunoassay [54].

• Data Realignment: Use fertility monitor data and serum hormone levels to reclassify clinic visits to the correct menstrual cycle phase based on biological markers rather than predetermined visit schedules [54].

• Multiple Imputation: Apply longitudinal multiple imputation methods to estimate hormone levels for missing cycle visits resulting from the realignment process [54].

• Validation: Compare realigned hormone profiles with expected physiological patterns. Realigned cycles should demonstrate more clearly defined hormonal profiles with higher mean peak hormones (up to 141%) and reduced variability (up to 71%) [54].

Applications: This protocol is particularly valuable for studying phase-specific associations, such as the relationship between daily fiber intake and reproductive hormone levels across different cycle phases [54].

Diagram: Workflow for menstrual cycle data realignment and imputation

Advanced Tracking Technologies and Machine Learning Approaches

Wearable Sensors and Machine Learning Protocol

Recent technological advances enable continuous physiological monitoring that can complement traditional hormone measurements for detecting menstrual cycle phases and imputing missing data.

Experimental Protocol: Machine Learning for Cycle Phase Classification

Objective: To classify menstrual cycle phases and detect ovulation using sleeping heart rate and machine learning, providing an alternative data stream for imputing missing phase information.

Materials:

• Wearable heart rate monitor (e.g., Fitbit Sense, Apple Watch)
• Machine learning platform (Python with XGBoost library)
• Optional: Basal body temperature thermometer
• Optional: Urinary luteinizing hormone tests

Procedure:

• Data Collection: Collect sleeping heart rate data continuously using wearable sensors. Focus on heart rate at the circadian rhythm nadir (minHR) as a key feature [18].

• Feature Engineering: Create three feature combinations for model evaluation:

  • "day": Number of days since menstruation onset
  • "day + minHR": Day count plus circadian rhythm nadir heart rate
  • "day + BBT": Day count plus basal body temperature [18]

• Model Development: Train an XGBoost machine learning model using nested leave-one-group-out cross-validation to classify menstrual cycle phases and predict ovulation day [18].

• Stratification: Stratify participants based on variability in sleep timing (high variability vs. low variability groups) [18].

• Performance Validation: Compare model performance across feature sets. The "day + minHR" model significantly improves luteal phase recall and reduces ovulation day detection absolute errors by 2 days compared to BBT-based models in participants with high sleep timing variability [18].

• Imputation Application: Use the trained model to predict cycle phases and ovulation timing in cases where direct hormone measurement data is missing.

Advantages: This approach is particularly robust for individuals with high variability in sleep timing, where traditional BBT methods perform poorly [18].

Multimodal Data Integration

The mcPHASES dataset exemplifies comprehensive multimodal data collection for menstrual health research, incorporating:

• Smartwatch-derived physiological metrics (heart rate, temperature, sleep quality, activity)
• Continuous glucose monitoring (Dexcom G6 CGM)
• Daily symptom diaries
• Daily hormone measurements from urinalysis (Mira Plus Starter Kit) [55]

This rich multimodal data enables researchers to develop more accurate imputation models by establishing relationships between easily measured parameters (e.g., heart rate) and gold-standard hormone measurements.

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 2: Research Reagent Solutions for Menstrual Cycle Studies

Item	Function	Application Context
Clearblue Easy Fertility Monitor	Measures urinary estrone-3-glucuronide and LH to predict ovulation	Timing clinic visits and realigning cycle phase classification [54]
Mira Plus Starter Kit	Quantitative urinary hormone analyzer for LH, E3G (estrogen), and PdG (progesterone)	Ground truth hormone measurement for validating other biomarkers [55]
DPC Immulite 2000 Analyzer	Solid phase competitive chemiluminescent enzymatic immunoassay for serum hormones	Gold standard measurement of serum estradiol, progesterone, LH, and FSH [54]
Fitbit Sense Smartwatch	Continuous monitoring of heart rate, temperature, sleep, and activity	Passive physiological data collection for machine learning models [55]
Dexcom G6 CGM	Continuous glucose monitoring	Investigating relationships between metabolic function and menstrual cycle [55]
Carolina Premenstrual Assessment Scoring System (C-PASS)	Standardized system for diagnosing PMDD and PME based on daily symptom ratings	Identifying and controlling for hormone-sensitive individuals in study samples [6]

Effective handling of missing data in menstrual cycle research requires specialized methodologies that account for the inherent biological variability of cycles and the within-person nature of cyclical changes. The protocols outlined in this document—from hormonal data realignment with multiple imputation to machine learning approaches using wearable sensor data—provide researchers with robust tools for maintaining data integrity.

As technological advances continue to expand opportunities for passive physiological monitoring, integrating these multimodal data streams will further enhance our ability to accurately impute missing menstrual cycle data. This progress will support more reliable phase-specific analyses in both basic research and clinical drug development, ultimately advancing women's health science.

Application Notes: Population-Specific Challenges and Algorithmic Adaptations

The development of robust menstrual cycle tracking algorithms requires specialized optimization to address the unique physiological patterns and health considerations of distinct populations. The following application notes detail the key challenges and corresponding algorithmic strategies for female athletes, individuals with Polycystic Ovary Syndrome (PCOS), and those in perimenopause.

Table 1: Population-Specific Challenges and Algorithmic Solutions

Population	Key Physiological Challenges	Proposed Algorithmic Adaptations	Primary Data Inputs
Athletes	High cycle length variability, impact of training load/intensity, risk of menstrual dysfunction [57] [58].	State-space models with overdispersion parameters, integration of covariate data (e.g., injury, training load) [57].	Cycle length history, self-reported symptoms (e.g., cramps, flow), training load metrics [57] [59].
PCOS	Menstrual irregularity (oligo-ovulation/anovulation), hyperandrogenism, polycystic ovarian morphology [60] [61].	Machine learning classification (e.g., XGBoost) using clinical, biochemical, and ultrasound features [60] [61].	Menstrual cycle history, BMI, hormone levels (LH, FSH, Testosterone, AMH), ovarian follicle count [60] [61].
Perimenopause	Increasing cycle length variability, anovulatory cycles, fluctuating and ultimately declining hormone levels [62] [63].	Detection of anovulation trends, stage classification algorithms (e.g., STRAW criteria), quantitative hormone level analysis [62] [63].	Quantitative E3G, LH, FSH, and PdG levels; cycle length patterns; symptom reports [62] [63].

Notes on Data Integration and Model Performance

• For Athletic Populations: A hybrid state-space model demonstrated high prediction accuracy for cycle length, with a root mean square error (RMSE) of 1.64 days and an overall accuracy of 0.9871 [57]. The model explicitly accounted for overdispersed cycles, which constituted 26.36% of cycles in the athletic sample [57].
• For PCOS Diagnosis: Machine learning models, particularly XGBoost, have achieved exceptional performance. One model utilizing combined clinical and ultrasound features reported an AUC of 0.9852, a precision of 0.9583, and an accuracy of 0.9384 [61].
• For Perimenopause Tracking: Quantitative hormone monitors (e.g., MIRA) track hormones like Estrone-3-Glucuronide (E3G) and Pregnanediol Glucuronide (PdG) to identify perimenopausal hormone patterns and confirm ovulatory status, which is crucial for accurate staging [63].

Experimental Protocols

Protocol 1: State-Space Modeling for Menstrual Cycle Length in Athletes

This protocol outlines the procedure for developing a predictive model for menstrual cycle length in athletes, as described in Scientific Reports (2021) [57].

Objective

To build a hybrid predictive model that captures within-subject temporal correlation and predicts the duration of an athlete's next menstrual cycle with high precision.

Materials and Dataset

• Data Source: Retrospective data from a menstrual cycle tracking application (e.g., FitrWoman).
• Cohort: Data from 2,125 women (age range: 18.00–47.10 years), comprising 16,524 menstrual cycles.
• Variables:
  • Primary Outcome: Menstrual cycle length (in days).
  • Covariates: Age, injury, stomach cramps, flow amount, tender breasts, training intensity.

Modeling Procedure

The modeling procedure is implemented in three sequential steps:

• Time Trend Component: Model the underlying temporal trend using a random walk process. Incorporate an overdispersion parameter (r_ij) to account for cycles with abnormally high variance.
• Autocorrelation Component: Capture serial dependence in the time series data using an autoregressive moving-average (ARMA) model. A moving average (MA(1)) parameter is often necessary to model the dynamism of shorter cycles followed by longer ones and vice versa.
• Linear Predictor: Integrate the effects of covariates (e.g., injury, stomach cramps) via a linear model to explain additional variation in cycle length.

Model Validation

• Testing Method: Use each woman's last observed cycle as a hold-out test set.
• Performance Metrics:
  • Root Mean Square Error (RMSE)
  • Concordance Correlation Coefficient (CCC)
  • Pearson Correlation Coefficient (r)

Protocol 2: Machine Learning for PCOS Diagnosis from Clinical Features

This protocol details the development of a non-invasive PCOS diagnostic model using machine learning, based on Scientific Reports (2025) [61].

Objective

To train and validate a machine learning model (XGBoost) for diagnosing PCOS based on the Rotterdam criteria, utilizing a combination of clinical and ultrasound features.

Materials and Dataset

• Data Source: Clinical dataset containing features aligned with Rotterdam criteria.
• Cohort: Patient records with confirmed PCOS and non-PCOS controls.
• Feature Categories:
  • Clinical: Menstrual irregularity, weight gain, hirsutism (hair growth), acne (pimples), hair loss.
  • Biochemical: Anti-Müllerian Hormone (AMH), luteinizing hormone (LH), follicle-stimulating hormone (FSH).
  • Ultrasound (USG): Bilateral ovarian follicle count.

Experimental Workflow

• Data Preprocessing: Handle missing values, normalize numerical features.
• Feature Selection: Apply the chi-square-based SelectKBest method to identify the top 10 most predictive features. Validate selection using XGBoost's internal feature importance and SHAP analysis.
• Model Training: Train an XGBoost classifier using various feature set combinations (e.g., Clinical + USG, Clinical + USG + AMH).
• Model Evaluation: Assess model performance on a held-out test set or via cross-validation.

Performance Metrics

• Area Under the Receiver Operating Characteristic Curve (AUC)
• Precision
• F1-Score
• Accuracy

Table 2: Key Predictive Features for PCOS Machine Learning Models

Feature Category	Specific Feature	Functional Role in Diagnosis
Ultrasound (USG)	Follicle count on both ovaries	Direct assessment of polycystic ovarian morphology (PCOM) [61].
Clinical	Weight gain / BMI (Obesity)	Marker of metabolic dysfunction commonly associated with PCOS [60] [61].
Biochemical	Anti-Müllerian Hormone (AMH)	Elevated levels are a surrogate marker for increased antral follicle count [61].
Clinical	Hair growth (Hirsutism)	Indicator of clinical hyperandrogenism [61].
Clinical	Menstrual irregularity	Indicator of ovulatory dysfunction [60] [61].
Clinical	Pimples (Acne)	Indicator of clinical hyperandrogenism [61].
Clinical	Hair loss	Indicator of clinical hyperandrogenism [61].
Biochemical	Luteinizing Hormone (LH)	Often elevated relative to FSH in PCOS [60].

Protocol 3: Quantitative Hormone Tracking for Perimenopause Staging

This protocol describes the use of a quantitative hormone monitor to track cycle characteristics and support the staging of perimenopause, as explored in Medicina (2023) [63].

Objective

To characterize hormonal cycle patterns during perimenopause using quantitative urinary hormone measurements to identify the fertile window and anovulatory cycles.

Materials

• Primary Device: Quantitative hormone monitor (e.g., MIRA monitor).
• Consumables: Single-use assay cartridges for urine testing.
• Software: Accompanying mobile application for data logging and visualization.

Procedure

• Data Collection: Participants provide daily first-morning urine samples throughout the study period.
• Hormone Measurement: The monitor quantitatively measures:
  • Estrone-3-Glucuronide (E3G): A urinary metabolite of estrogen.
  • Luteinizing Hormone (LH).
  • Follicle-Stimulating Hormone (FSH).
  • Pregnanediol Glucuronide (PdG): A urinary metabolite of progesterone.
• Data Analysis:
  • Fertile Window Initiation: In regular cycles, an E3G level of >100 ng/mL is proposed to mark the beginning of the fertile window. In perimenopause, fluctuations are common, and a protocol of E3G ≥ 100 ng/mL for one or two days may be used.
  • Ovulation Confirmation: An LH surge (e.g., ≥11 mIU/mL) is used to predict ovulation.
  • Ovulation Verification: A sustained rise in PdG is used to confirm that ovulation has likely occurred.
• Cycle Classification: Cycles are categorized based on STRAW criteria (e.g., Regular, Short, Variable, >60 days without cycles) using the collected hormonal and cycle length data.

Visualization of Workflows and Relationships

Menstrual Cycle Tracking Model Development

PCOS Diagnostic Model Workflow

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Tools for Menstrual Cycle Research

Tool / Reagent	Function / Application	Example / Notes
State-Space Model	Statistical modeling for time-series data with high within-subject variability.	Used for predicting cycle length in athletes; incorporates random walk and overdispersion parameters [57].
XGBoost Classifier	A machine learning algorithm for classification and regression tasks.	Preferred for PCOS prediction models due to high performance and feature importance output [61].
Quantitative Hormone Monitor	Device for measuring precise concentrations of reproductive hormones in urine.	MIRA monitor measures E3G, LH, FSH, PdG; crucial for perimenopause research [63].
Basal Body Thermometer	High-precision thermometer for tracking subtle basal body temperature shifts.	Required for temperature-based algorithms; must measure to two decimal places [62].
SelectKBest (χ²)	Feature selection method to identify the most relevant predictors from a dataset.	Used in PCOS research to rank features like follicle count and AMH [61].
SHAP (SHapley Additive exPlanations)	A method to interpret the output of machine learning models.	Explains the contribution of each feature to an individual PCOS prediction [61].
Clearblue Monitor	Qualitative urinary hormone monitor providing threshold-based readings.	An alternative tool for fertility tracking; measures estrogen and LH metabolites [63].

Incorporating Privacy-Preserving Techniques like Federated Learning

The integration of artificial intelligence (AI) into women's health, particularly for menstrual cycle and ovulation tracking, represents a rapidly advancing field. However, this progress is accompanied by significant privacy concerns. Conventional digital health applications often rely on centralized data storage and processing models, where sensitive user information, including menstrual cycle dates, symptoms, and sexual activity, is transferred to developer servers. This practice creates substantial privacy risks, as this highly intimate data can be vulnerable to security breaches, unauthorized sharing with third parties for advertising, and compelled disclosure to law enforcement, particularly in jurisdictions where reproductive rights are under threat [64] [65] [66]. In this context, privacy-preserving computational techniques like Federated Learning (FL) have emerged as a foundational technology for developing responsible and trustworthy digital health tools. This document outlines application notes and experimental protocols for incorporating FL into research on coding menstrual cycle day phase protocols, providing a framework that aligns with the demands of both scientific rigor and data ethics.

Technical Foundations of Federated Learning for Health Data

Federated Learning is a distributed machine learning approach that enables model training across multiple decentralized devices or servers holding local data samples, without exchanging them [67]. This paradigm is particularly suited for sensitive health data as it operates on a fundamental principle: data remains on the local device.

Core Principle and Workflow

In a typical FL system for menstrual health tracking, a global predictive model is trained as follows:

• A central server initializes a global machine learning model (e.g., for ovulation prediction) and distributes it to user devices or client servers.
• Each client device trains the model locally using its own private data (e.g., physiological signals and cycle logs).
• Instead of raw data, only the computed model updates (e.g., gradients or weights) are sent back to the central server.
• The server aggregates these updates from many devices to improve the global model.
• The updated global model is then redistributed, and the process repeats [68] [67].

This process ensures that sensitive reproductive health data never leaves the user's device, thereby minimizing the risk of privacy breaches and unauthorized data access [69].

Synergy with Other Privacy-Enhancing Technologies (PETs)

To further bolster security, FL can be integrated with other PETs, creating a multi-layered privacy defense:

• Differential Privacy (DP): This technique adds a carefully calibrated amount of statistical noise to the model updates before they are shared. This noise obscures the contribution of any single individual, making it virtually impossible to reverse-engineer or identify a specific user's data from the aggregated model, thus providing a strong, mathematical guarantee of privacy [69].
• Fully Homomorphic Encryption (FHE): FHE allows computations to be performed directly on encrypted data. In an FL context, clients could send their model updates in an encrypted form. The server could then aggregate these encrypted updates without needing to decrypt them, offering an additional layer of protection during transmission and aggregation [69].
• Blockchain Technology: Blockchain can be utilized to enhance transparency and trust in the FL process. It can provide an immutable, tamper-proof ledger for tracking user consent preferences and recording all global model updates, ensuring the integrity and auditability of the entire training process [69].

Table 1: Comparison of Privacy-Enhancing Technologies (PETs) for Federated Learning Systems

Technology	Primary Function	Key Advantage	Potential Drawback
Federated Learning (FL)	Decentralized model training; data remains on-user device.	Prevents raw data collection; mitigates breach risk.	Model updates might still leak information.
Differential Privacy (DP)	Adds statistical noise to data or model outputs.	Provides a mathematical privacy guarantee.	Can slightly reduce model accuracy.
Fully Homomorphic Encryption (FHE)	Enables computation on encrypted data.	Protects data during processing and aggregation.	Computationally intensive; can slow training.
Blockchain	Provides decentralized, immutable record-keeping.	Ensures transparency and auditability of model updates.	Can introduce scalability and complexity challenges.

Proposed Architecture and Implementation

A robust FL system for menstrual health research requires a carefully designed architecture that addresses privacy, functionality, and practical deployment constraints.

System Architecture and Data Flow

The following diagram illustrates the flow of data and models in a privacy-preserving Federated Learning system for menstrual health research.

This architecture ensures a closed loop where the global model improves without centralizing sensitive raw data.

Experimental Protocol for FL Model Development

This protocol provides a step-by-step guide for researchers to develop and validate a federated learning model for menstrual phase prediction.

Objective: To collaboratively train a machine learning model that predicts menstrual cycle phases (e.g., follicular, ovulatory, luteal) using decentralized physiological data without centralizing raw user information.

Materials: The "Scientist's Toolkit" in Section 5 lists essential reagents and computational tools.

Method:

• Problem Formulation & Model Selection:
  • Define Prediction Task: Clearly specify the target variable (e.g., next-day phase: menstruation, pre-ovulation, ovulation, post-ovulation).
  • Select Base Model Architecture: Choose a model suitable for time-series and classification tasks. Research indicates Random Forest classifiers can achieve up to 91% accuracy in predicting three menstrual phases, while Long Short-Term Memory (LSTM) networks also show high promise due to their ability to model temporal sequences [70].

• Federated Training Setup:

  • Initialize Global Model: The central server initializes the chosen model with random weights.
  • Client Selection: In each training round, a subset of clients (user devices or institutional servers) is selected. Strategies can be random or resource-aware [71].
  • Distribution: The current global model is distributed to all selected clients.

• Local Training on Client Devices:

  • Data Preprocessing: Each client preprocesses its local data. For physiological signals like wrist skin temperature and heart rate, this may involve cleaning and normalization. One study achieved a Root Mean Square Error (RMSE) of 0.133 ± 0.055 °C when predicting next-day skin temperature using a Random Forest model [70].
  • Model Training: Each client trains the received model on its local dataset for a predetermined number of epochs.
  • Update Preparation: The trained model's weights/gradients are prepared for transmission. At this stage, Differential Privacy noise can be added to the updates to enhance privacy [69].

• Aggregation and Model Update:

  • Secure Transmission: Encrypted model updates are sent to the central server.
  • Server-Side Aggregation: The server aggregates the updates using an algorithm like Federated Averaging (FedAvg) to create a new, improved global model [68] [67].
  • Validation: The updated global model can be evaluated on a held-out test set or by requesting validation metrics from clients without sharing their data.

• Iteration and Evaluation:

  • Steps 2-4 are repeated for multiple communication rounds until the model converges to a satisfactory performance level.
  • The final model is evaluated for both predictive performance (e.g., accuracy, AUC-ROC) and privacy guarantees (e.g., the epsilon value in Differential Privacy).

Data Handling and Computational Validation

A critical aspect of FL system design involves managing heterogeneous data and validating the model in conditions that mimic real-world deployments.

Handling Statistical Heterogeneity

A key challenge in FL is that data across clients is typically non-independently and identically distributed (non-IID). User cycle patterns, physiological responses, and lifestyle factors vary significantly. To emulate this realistically in research:

• Use Realistic Data Partitions: Utilize frameworks like the LEAF benchmark or Flower Datasets library, which provide tools to partition datasets in a non-IID fashion based on real-world patterns [71].
• Partitioning Strategies: Implement shard-based partitioning (assigning different class labels to different clients) or use a Dirichlet distribution to vary class concentrations across clients, thereby simulating the natural diversity of menstrual cycle characteristics within a population [71].

Performance and Validation Metrics

To holistically evaluate the FL system, researchers should track the following metrics, which correlate algorithmic performance with system constraints:

Table 2: Key Validation Metrics for Federated Menstrual Cycle Prediction Models

Metric Category	Specific Metric	Target Value/Benchmark	Justification
Predictive Accuracy	Phase Prediction Accuracy	>88-91% [70]	Accuracy achieved by Random Forest & LSTM models in research settings.
	Area Under the Curve (AUC)	>0.92 [67]	AUC achieved by a federated model (EXAM) in clinical outcome prediction, demonstrating FL's potential for high performance.
Model Generalizability	Performance Variation Across Client Sites	<16% degradation vs. local models [67]	Federated models can show a 16% improvement in AUC and 38% increase in generalizability compared to models trained on a single site's data.
Privacy & Efficiency	Differential Privacy Epsilon (ε)	A lower value (e.g., ε < 5) indicates stronger privacy.	A key parameter for quantifying the privacy-utility trade-off.
	Communication Rounds	Minimized for convergence	Reduces overall training time and resource usage.
	Local Computational Load	Monitor CPU/Memory usage on client devices.	Ensures feasibility of on-device training without degrading user experience.

The following diagram outlines the workflow for the experimental protocol, from data collection to model validation, highlighting key decision points.

The Scientist's Toolkit

This section details the essential materials, software, and data sources required to implement the proposed protocols.

Table 3: Research Reagent Solutions for Federated Learning Experiments

Item Name / Category	Specifications / Examples	Function / Application in Protocol
Federated Learning Frameworks	Flower, TensorFlow Federated, PySyft	Provides the core software infrastructure for orchestrating the federated learning process, including communication, aggregation, and client management.
Network Emulation Testbeds	FLEET, MininetFed	Enables high-fidelity emulation of real-world network conditions (bandwidth, latency) to test FL system robustness and efficiency before deployment [71].
Privacy-Enhancing Technologies (PETs)	Differential Privacy Libraries (e.g., TensorFlow Privacy), Fully Homomorphic Encryption Libraries (e.g., Microsoft SEAL)	Implements mathematical privacy guarantees by adding noise to model updates (DP) or enabling computation on encrypted data (FHE) [69].
Machine Learning Models & Datasets	Random Forest, LSTM; Datasets with physiological signals (wrist temperature, HR, IBI)	Serves as the predictive algorithm and training data. Random Forest has shown 91% accuracy for phase prediction; LSTMs are effective for time-series data [70].
Data Partitioning Tools	LEAF Benchmark, Flower Datasets Library	Simulates realistic, non-IID data distributions across clients, which is crucial for evaluating the generalizability of the federated model [71].
Blockchain Platform	Ethereum, Hyperledger Fabric	(Optional) Provides an immutable ledger for tracking model versioning and user consent, enhancing transparency and trust in the system [69].

Validation and Comparative Analysis: Establishing Methodological Credibility

Accurate determination of the luteinizing hormone (LH) surge and concomitant hormonal fluctuations is fundamental to reproductive biology research, particularly in studies investigating menstrual cycle phase effects on physiological parameters. The complex hormonal interactions between pituitary and ovarian hormones regulate follicular development, ovulation, and endometrial preparation [72]. For research aimed at establishing menstrual cycle phase protocols, implementing gold standard validation methods for LH surge confirmation and hormonal profiling is methodologically critical. This application note outlines evidence-based protocols and analytical considerations for robust hormonal assessment in research settings, with particular emphasis on addressing common methodological challenges in phase determination.

Gold Standard Methodologies for Hormonal Assessment

The validation of menstrual cycle phases in research requires a multi-factorial approach that combines direct hormonal measurements with physiological observations. Transvaginal ultrasound for tracking follicular development and serum hormone testing for estradiol (E2), progesterone (P4), and luteinizing hormone (LH) are widely recognized as the clinical and research gold standards [73]. These methods provide the most accurate and reliable data for pinpointing ovulation and defining hormonally discrete cycle phases.

However, practical constraints in research settings have led to the development and validation of alternative methods. The table below summarizes the key methodologies for hormonal assessment and ovulation detection, their applications, and limitations.

Table 1: Comparison of Methodologies for Hormonal Assessment and Ovulation Detection

Methodology	Primary Application	Key Measures	Validity & Precision Considerations
Serum Hormone Assays	Gold standard for phase confirmation [73]	Quantitative LH, E2, P4	High sensitivity and specificity; requires venipuncture
Transvaginal Ultrasound	Gold standard for visualizing ovulation [73]	Follicle size, endometrial thickness	Direct observation of follicular rupture; operator-dependent
Quantitative Urinary LH Monitors	At-home LH surge detection [74] [75]	Urinary LH metabolites	High correlation with serum LH surge [74]; identifies fertile window
Urinary E1G Testing	Tracking estrogen rise [74]	Estrone-3-glucuronide (E1G)	Correlates with serum estradiol; defines beginning of fertile window
Salivary Hormone Assays	Field-based progesterone assessment [73]	Salivary E2 and P4	Measures bioavailable hormone; variable validity and precision

Experimental Protocols for LH Surge Confirmation and Cycle Phase Validation

Protocol 1: Serum-Based LH Surge and Hormonal Phase Determination

This protocol is designed for laboratory-based research requiring high-precision phase identification.

Materials and Reagents:

• EDTA or serum separation tubes
• Automated immunoassay platform or ELISA kits for LH, E2, P4
• Centrifuge
• -20°C freezer for sample storage

Procedure:

• Baseline Assessment: Draw blood on cycle day 2-3 to establish baseline FSH and E2 levels in participants with confirmed cycle lengths of 21-35 days.
• High-Frequency Sampling During Ovulatory Window: Begin daily blood draws when a dominant follicle reaches ≥16mm via ultrasound or when urinary LH tests indicate rising levels.
• LH Surge Confirmation: The day of the LH surge is defined as the first day when serum LH concentration rises ≥50% above the average of the previous five days and exceeds an absolute threshold (typically >15-25 mIU/mL) [76].
• Ovulation Confirmation: A sustained rise in serum progesterone to >5 ng/mL approximately 3-5 days post-LH surge confirms ovulation and luteal phase function [5].
• Phase Definition:
  • Follicular Phase: From menses until the day before the LH surge.
  • Ovulatory Phase: The day of the LH surge and the following 24 hours.
  • Luteal Phase: From 2 days post-LH surge until the onset of subsequent menses, with mid-luteal P4 >10 ng/mL indicating robust luteal function.

Protocol 2: Urinary Hormone Monitoring for Field-Based Research

This protocol validates the use of quantitative urinary hormone monitors for LH surge detection in non-laboratory settings.

Materials and Reagents:

• Quantitative urinary hormone monitor (e.g., Mira Monitor, ClearBlue Fertility Monitor)
• Corresponding test strips for LH and E1G
• Sterile urine collection cups
• First-morning urine samples

Procedure:

• Cycle Initiation: Participants begin testing on cycle day 6 or according to device instructions.
• Daily Testing: Collect first-morning urine daily. Test samples immediately upon collection using the quantitative monitor.
• LH Surge Identification: The urinary LH surge is defined as the first day when LH values exceed a predefined threshold (e.g., ≥17 mIU/mL [76] or laboratory-defined cutoff). Studies show a high correlation (R=0.94) between the LH surge detected by quantitative monitors and established fertility monitors [74].
• Estrogen Rise Monitoring: Concurrently monitor E1G levels. A sustained rise indicates follicular development and the beginning of the fertile window.
• Methodological Validation: For research purposes, a subset of participants should undergo concurrent serum testing to validate urinary hormone measures in the specific study population.

The following workflow diagram illustrates the decision-making process for integrating these methodologies in a research setting.

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Research Reagents and Materials for Menstrual Cycle Hormone Detection

Item	Function/Application	Specification Considerations
LH Immunoassay Kits	Quantifying LH in serum/urine	Detect intact LH molecule; sensitivity <0.5 mIU/mL; cross-reactivity with hCG <1% [75]
Progesterone ELISA	confirming ovulation & luteal function	Specific for P4; report dynamic range (e.g., 0.3-60 ng/mL); intra-assay CV <10%
Estradiol RIAs/ELISAs	tracking follicular development	Sensitivity <10 pg/mL; minimal cross-reactivity with estrone
Urinary LH Test Strips	detecting LH surge in field settings	Qualitative or quantitative; threshold ~20-40 mIU/mL; >99% detection of LH surge [74]
Salivary Collection Kits	non-invasive progesterone monitoring	Use salivettes with cotton or polyester rolls; assess intra-assay CV [73]
Microfluidic Biosensors	quantitative, rapid LH detection	Electrochemical impedance detection; LOD ~1.0 mIU/mL; agitation enhances signal [75]

Analytical Considerations and Methodological Challenges

Assay Validation and Quality Control

Robust hormonal assessment requires rigorous validation of all analytical methods. Researchers must report key assay quality parameters, including:

• Sensitivity (limit of detection)
• Specificity (minimal cross-reactivity with similar hormones)
• Precision (intra- and inter-assay coefficients of variation)

Salivary and urinary methods show promise for field-based studies but require careful validation against serum standards due to reported inconsistencies in validity and precision [73]. For instance, salivary assays measure the bioavailable fraction of hormones, while urinary assays detect hormone metabolites, leading to potential discrepancies with serum values [73].

Critical Importance of Direct Measurement

A significant concern in menstrual cycle research is the practice of assuming or estimating cycle phases without direct hormonal measurement. This approach lacks scientific rigor and can lead to misclassification of cycle phases [5]. Calendar-based counting alone cannot detect anovulatory cycles or luteal phase deficiencies, which are common in athletic populations [5]. Direct measurement of the LH surge and subsequent progesterone rise is essential for valid phase classification in research contexts.

Implementing gold standard methodologies for LH surge confirmation and hormonal assays is imperative for generating valid, reliable data in menstrual cycle research. While serum testing remains the benchmark, validated urinary hormone monitors and emerging biosensor technologies offer practical alternatives for field-based studies. Critical methodological considerations include rigorous assay validation, appropriate sampling frequency, and direct measurement of key hormonal events rather than calendar-based estimates. By adhering to these detailed protocols and analytical standards, researchers can significantly enhance the quality and reproducibility of studies investigating menstrual cycle phase effects.

In the realm of computational research, particularly in developing classification models for menstrual cycle phase prediction, the rigorous evaluation of model performance is paramount. The selection of appropriate metrics directly impacts the interpretability, reliability, and clinical applicability of research findings. For researchers and drug development professionals working with physiological data, understanding the trade-offs and interpretations of these metrics ensures that developed models are not only statistically sound but also clinically meaningful.

The four fundamental metrics—Accuracy, Precision, Recall, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC)—serve distinct purposes in model assessment. Accuracy provides an overall measure of correct predictions, while Precision quantifies the reliability of positive predictions, and Recall (also known as Sensitivity) measures the ability to identify all relevant instances. The AUC-ROC curve offers a comprehensive view of model performance across all classification thresholds, balancing the true positive rate against the false positive rate [77].

In menstrual cycle research, where phase classification can inform fertility treatments, hormonal disorder diagnosis, and drug development protocols, these metrics help validate models against clinical standards. For instance, in detecting the ovulation phase, high Recall is often prioritized to minimize false negatives, whereas for luteal phase identification, Precision might be more critical to avoid misclassifying non-luteal phases [18] [11].

Metric Definitions and Computational Methods

Mathematical Foundations and Formulas

Each performance metric offers a unique perspective on model behavior by leveraging different components of the confusion matrix: True Positives (TP), True Negatives (TN), False Positives (FP), and False Negatives (FN).

Accuracy measures the overall correctness of the model across all classes, calculated as the ratio of correct predictions to total predictions: ( \text{Accuracy} = \frac{TP + TN}{TP + TN + FP + FN} ). While easily interpretable, accuracy can be misleading with imbalanced class distributions, common in medical datasets where one class may be underrepresented [77].

Precision evaluates the exactness of positive predictions, reflecting how often a model is correct when it predicts a positive class: ( \text{Precision} = \frac{TP}{TP + FP} ). In menstrual cycle phase classification, high precision for the ovulation phase means that when the model predicts ovulation, it is highly likely to be correct, reducing false alarms [77].

Recall (Sensitivity) assesses the model's ability to capture all positive instances, measuring completeness: ( \text{Recall} = \frac{TP}{TP + FN} ). High recall in menstrual phase detection ensures that actual ovulation or menses events are not missed, which is crucial for fertility and health applications [77].

AUC-ROC represents the model's ability to distinguish between classes across all possible thresholds. The ROC curve plots the True Positive Rate (Recall) against the False Positive Rate (FPR = ( \frac{FP}{FP + TN} )) at various threshold settings. The area under this curve provides a single scalar value between 0.5 (random guessing) and 1.0 (perfect classifier) [77].

Practical Computation Using Python

Implementation of these metrics is straightforward using common data science libraries. The following code demonstrates calculation using scikit-learn:

Output:

For multi-class menstrual phase classification (e.g., menstruation, follicular, ovulation, luteal), metrics can be calculated using micro, macro, or weighted averaging to account for class imbalances, which are common in physiological data [11].

Application in Menstrual Cycle Phase Classification

Performance Metrics in Recent Menstrual Cycle Research

Recent studies applying machine learning to menstrual phase classification have demonstrated the critical importance of selecting appropriate evaluation metrics. The following table summarizes performance metrics reported in recent studies:

Table 1: Performance Metrics in Menstrual Cycle Phase Classification Studies

Study Reference	Classification Task	Model Used	Accuracy	Precision	Recall	AUC-ROC	Key Findings
Nature (2025) [11]	3-phase classification (Period, Ovulation, Luteal)	Random Forest	0.87	0.87	0.87	0.96	Fixed window feature extraction outperformed rolling windows
Nature (2025) [11]	4-phase classification (Period, Follicular, Ovulation, Luteal)	Random Forest	0.71	N/R	N/R	0.89	Increased phase complexity reduces performance
Scientific Reports (2025) [18]	Ovulation day detection	XGBoost	N/R	N/R	N/R	N/R	minHR feature reduced detection errors by 2 days in high sleep variability
Scientific Reports (2025) [18]	Luteal phase classification	XGBoost	N/R	Improved recall with minHR	Improved recall with minHR	N/R	minHR-based features outperformed BBT in recall

N/R = Not explicitly reported in the source material

Metric Selection for Specific Research Objectives

The choice of emphasis among metrics should align with the clinical or research application. For fertility-focused applications where missing an ovulation event has significant consequences (e.g., in natural family planning or conception timing), recall should be prioritized to minimize false negatives [18]. In contrast, for symptom management applications where phase-specific interventions are implemented (e.g., for premenstrual dysphoric disorder), precision may be more important to ensure interventions are only applied during correct phases [6].

The AUC-ROC is particularly valuable for comparing different models or feature sets in menstrual phase classification, as it provides a threshold-agnostic evaluation. For instance, in a study comparing heart rate-based features against traditional basal body temperature (BBT), the AUC-ROC can objectively demonstrate which modality provides better separation between menstrual phases independent of the specific threshold chosen for clinical implementation [18] [11].

Experimental Protocols for Metric Evaluation

Standardized Evaluation Workflow for Menstrual Phase Classification

Implementing a consistent evaluation protocol ensures comparable results across studies. The following workflow outlines a standardized approach for assessing performance metrics in menstrual cycle research:

Diagram 1: Experimental workflow for evaluating classification performance in menstrual cycle studies

Detailed Protocol for Cross-Validation in Menstrual Cycle Studies

Objective: To implement rigorous evaluation of classification metrics while accounting for within-subject and between-subject variability in menstrual cycle data.

Materials and Equipment:

• Physiological data collection devices (wearable sensors for HR, HRV, temperature)
• Data preprocessing software (Python, R, or specialized biosignal processing tools)
• Ground truth validation tools (LH test kits, BBT thermometers, symptom logs)
• Computing environment with machine learning libraries (scikit-learn, TensorFlow, PyTorch)

Procedure:

• Data Collection and Labeling

  • Collect physiological data (heart rate, heart rate variability, skin temperature) using wearable sensors at appropriate sampling frequencies [11]
  • Establish ground truth labels for menstrual phases using:
    • Urinary luteinizing hormone (LH) tests to pinpoint ovulation
    • Basal body temperature (BBT) tracking to confirm luteal phase
    • Menstrual bleeding logs to mark menstruation phase
    • Hormone assays (estradiol, progesterone) where feasible [6]

• Feature Engineering

  • Extract circadian rhythm-based features such as heart rate at circadian rhythm nadir (minHR) [18]
  • Calculate daily averages, variability metrics, and phase-specific trends
  • Normalize features to account for inter-individual differences

• Data Partitioning

  • Implement leave-one-subject-out or leave-last-cycle-out cross-validation to prevent data leakage and test generalizability [11]
  • Ensure proportional representation of all phases in training and test sets

• Model Training and Evaluation

  • Train multiple classifier types (Random Forest, XGBoost, Neural Networks)
  • Generate prediction probabilities for each phase
  • Calculate performance metrics (Accuracy, Precision, Recall, AUC-ROC) for each phase and overall
  • Compare performance across different feature sets and model architectures

Expected Outcomes: The protocol should yield reproducible metric evaluations that accurately reflect real-world performance, highlighting which models and features are most suitable for specific menstrual phase classification tasks.

Advanced Visualization Techniques

Comprehensive Metric Visualization Approaches

Effective visualization of performance metrics enhances interpretation and communication of results. The following diagram illustrates the relationship between different evaluation metrics and their visualization techniques:

Diagram 2: Visualization techniques and metric relationships for comprehensive model evaluation

Implementation of Visualization Code

The following Python code demonstrates generation of key visualizations for performance metrics:

Research Reagent Solutions and Essential Materials

Table 2: Essential Research Materials for Menstrual Cycle Phase Classification Studies

Category	Specific Tool/Reagent	Application in Research	Performance Consideration
Physiological Sensors	Wrist-worn devices (E4, EmbracePlus)	Continuous measurement of HR, HRV, EDA, temperature	Sampling frequency impacts feature quality [11]
Ground Truth Validation	Urinary LH test kits	Objective identification of ovulation day	Gold standard for ovulation timing [6]
Ground Truth Validation	Basal body temperature (BBT) thermometers	Confirmation of luteal phase and ovulation	High sensitivity to sleep disruptions [18]
Data Processing	Python/R with scikit-learn, TensorFlow	Model development and metric computation	Flexibility for custom metric implementation [77]
Hormonal Assays	ELISA kits for estradiol, progesterone	Hormonal correlation with physiological features	Provides biological validation but costly for large n [6]
Data Collection Platforms	Mobile health applications	Symptom logging, cycle tracking	Enables large-scale data collection [78]

The rigorous application of performance metrics—Accuracy, Precision, Recall, and AUC-ROC—is essential for advancing the field of menstrual cycle phase classification using computational methods. By implementing standardized evaluation protocols, selecting metrics aligned with research objectives, and utilizing appropriate visualization techniques, researchers can develop more reliable and clinically applicable models. The integration of diverse data sources, from wearable sensors to hormonal assays, coupled with thoughtful metric selection, will continue to enhance our understanding of female physiology and contribute to improved health outcomes through personalized medicine approaches.

Comparative Analysis of Model Performance Across Different Data Modalities

Within the burgeoning field of female athlete research, the integration of diverse data types is paramount for developing a holistic understanding of how the menstrual cycle (MC) influences performance and recovery. The broader thesis of this work posits that robust, phase-specific coding protocols are foundational for generating comparable and actionable scientific insights. This application note provides a detailed comparative analysis of model performance when built upon subjective, objective, and combined data modalities, and subsequently outlines standardized protocols for MC research to guide researchers and drug development professionals.

Quantitative Data Synthesis

The following tables synthesize key quantitative findings from recent literature, highlighting the relationships between MC phases, symptom burden, and various performance metrics.

Table 1: Menstrual Cycle Phase Definitions and Hormonal Profiles [79] [80] [81]

Phase	Approximate Days (from LMP)	Key Hormonal Characteristics	Reported Performance Trends
Early Follicular (EF)	1 - 5	Low estrogen, low progesterone	Perceived & objective performance often worst; strength & aerobic capacity may be reduced. [81]
Late Follicular (LF)	6 - 13	High estrogen, low progesterone	Often favorable for strength & power; potential for best performance. [80] [81]
Ovulatory (O)	~14	Peak estrogen, LH surge	Mixed performance trends; some report best anaerobic & strength performance. [81]
Early-to-Mid Luteal (EL/ML)	15 - 24	High progesterone, moderate estrogen	Increased perceived exertion & thermoregulatory strain; endurance may be impaired. [79] [81]
Late Luteal (LL)	25 - 28+	Declining progesterone & estrogen	High symptom burden; perceived performance low; strength & aerobic output may be worst. [79] [81]

Table 2: Comparative Impact of Cycle Phase vs. Symptom Burden on Key Outcomes (Synthesis of Elite Athlete Studies) [79]

Factor	Sleep Quality	Recovery State	Stress State	Overall Performance
Menstrual Cycle Phase	Limited & inconsistent associations	Limited & inconsistent associations	Limited & inconsistent associations	Inconsistent objective results; ~57% of studies show no difference. [81]
Daily Symptom Burden	Consistently associated with poorer quality	Consistently associated with reduced recovery	Consistently associated with elevated stress	>50% of athletes report perceived impairment. [79] [81]
Key Findings	Symptom burden is a more relevant factor than hormonal phase for sleep and recovery. [79]	Individual variability is high; personalized monitoring is crucial. [79] [80]

Experimental Protocols for Menstrual Cycle Research

Protocol 1: Longitudinal Monitoring of Sleep and Recovery-Stress States

This protocol is adapted from a study on elite female basketball players. [79]

• Aim: To examine the influence of MC phases and daily symptom burden on sleep quality and recovery-stress states.
• Subject Selection:
  • Inclusion: Healthy, eumenorrheic females with natural menstrual cycles (typical length 21-35 days). Athletes should be classified as Tier 3 (Highly Trained/National) or Tier 4 (Elite/International).
  • Exclusion: Use of hormonal contraception; presence of conditions like PCOS or amenorrhea.
• Data Collection Timeline: Minimum of one complete menstrual cycle (recommended 3 months for observational studies).
• Primary Data Modalities:
  • Objective Hormonal Tracking: Verify cycle phase and regularity using salivary hormone samples (e.g., estrogen, progesterone) collected twice weekly or using a certified fertility tracker (e.g., Ava bracelet).
  • Subjective Daily Diaries: Participants complete daily entries for:
    • Menstrual symptoms (e.g., fatigue, cramps, mood changes) and bleeding intensity.
    • Subjective sleep quality (e.g., using a Likert scale).
    • Recovery-stress state (e.g., using the Recovery-Stress Questionnaire for Athletes).
  • Objective Sleep Monitoring: Use wearable devices (e.g., actigraphy) to track sleep parameters like duration, wake after sleep onset (WASO), and sleep efficiency.
• Data Management & FAIRification:
  • Structure data in spreadsheets using a consistent format from the study's inception, similar to the ODAM (Open Data for Access and Mining) approach. [82]
  • Define all columns (e.g., participant_id, cycle_day, symptom_score, sleep_quality) unambiguously.
  • Use unique, persistent identifiers for participants and link data tables via these IDs to facilitate combination and merging for analysis.
  • Upon study completion, deposit the structured dataset in an appropriate repository using a standard like Frictionless Data Package to enhance findability, accessibility, interoperability, and reusability (FAIR). [82]
• Statistical Analysis:
  • Use Linear Mixed Models (LMM) to account for repeated measures and intra-individual variation. [79]
  • Model fixed effects (e.g., cycle phase, symptom burden) and random effects (e.g., participant ID).

Protocol 2: Qualitative Assessment of Perceived Performance

This protocol is adapted from a study on women's experiences in strength training. [80]

• Aim: To explore women's perceptions and experiences of performance (e.g., strength, motivation, energy) during different MC phases.
• Study Design: Qualitative, using a conventional content analysis with an inductive approach.
• Subject Selection:
  • Inclusion: Women (18-45 years) with recreational experience in strength training (≥2 times/week for ≥1 year) and who track their MC using basal body temperature (BBT) or a certified app (e.g., Natural Cycles).
  • Exclusion: Use of hormonal contraception.
• Procedure:
  • Training Diary: Participants maintain a diary for one complete MC, recording experiences after each training session regarding strength (volume, intensity), perceived motivation, and energy levels.
  • Semi-structured Interviews: Conduct 30-45 minute interviews in the participant's native language, guided by a script covering: (1) general strength training experiences, (2) MC experiences, and (3) the relationship between the MC and physical performance.
  • Data Processing: Interviews are transcribed verbatim and translated if necessary.
• Data Analysis:
  • Familiarization: Read transcripts multiple times to gain an overall impression.
  • Identifying Meaning Units: Locate the smallest segments of text relevant to the research question.
  • Coding: Condense meaning units and label them with codes.
  • Categorization: Group codes into categories and overarching themes that describe the participants' lived realities.

Visual Workflows and Logical Relationships

The following diagrams illustrate the core workflows and logical frameworks for the protocols described above.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Tools for Menstrual Cycle Research

Item / Solution	Function / Purpose	Example Specifications & Notes
Salivary Hormone Kits	Non-invasive collection of estrogen, progesterone, testosterone, and cortisol for objective phase verification and hormonal profiling.	Salivettes; require freezer storage; analyze via ELISA or LC-MS.
Basal Body Temperature (BBT) Thermometer	Tracking subtle shifts in resting body temperature to confirm ovulation and delineate follicular vs. luteal phases.	High-precision (2 decimal places) digital thermometers; used upon waking.
Validated Psychometric Questionnaires	Quantifying subjective experiences of symptoms, recovery, stress, and sleep quality.	Recovery-Stress Questionnaire (RESTQ), Pittsburgh Sleep Quality Index (PSQI), custom symptom diaries using Likert scales.
Activity/Sleep Wearables	Objective, continuous monitoring of sleep parameters (duration, efficiency, WASO) and activity load.	Actigraphy watches (e.g., ActiGraph), consumer devices (e.g., Garmin, Whoop); requires consistency of use.
Data Structuring Software	Implementing FAIR data principles from study inception; structuring and annotating complex longitudinal data for analysis.	Spreadsheet software (Excel, Google Sheets) with strict protocols; ODAM method; R or Python scripts for automated processing. [82]
Statistical Software with LMM & EL	Advanced statistical analysis accounting for repeated measures (LMM) and non-parametric data distributions (Empirical Likelihood).	R (lme4 package), Python (statsmodels), ILLMO software, SAS. [83]

Benchmarking Against Existing Commercial Apps and Clinical Methods

Accurately determining menstrual cycle phases is critical for clinical research, drug development, and women's health studies. Traditional methods for phase determination range from simple calendar-based counting to sophisticated hormonal assays, each with varying degrees of precision, practicality, and validation. For researchers and drug development professionals, selecting an appropriate benchmarking method requires careful consideration of accuracy, resource constraints, and specific research objectives. This document provides a comprehensive comparison of existing commercial applications and clinical methodologies, detailing their underlying mechanisms, performance metrics, and implementation protocols. The content is framed within the broader context of developing robust, code-based protocols for menstrual cycle phase classification in research settings, with emphasis on methodological rigor and valid outcome measurement.

Performance Benchmarking of Tracking Methods

The landscape of menstrual cycle tracking technologies encompasses traditional clinical methods, modern wearable-based algorithms, and commercial applications. The table below summarizes the key performance metrics and characteristics of these approaches, providing researchers with comparative data for methodological selection.

Table 1: Performance Metrics of Menstrual Cycle Tracking Technologies

Method / Technology	Underlying Data Inputs	Reported Accuracy / Performance	Key Advantages	Key Limitations
Machine Learning (XGBoost) with minHR [18]	Sleeping heart rate at circadian nadir (minHR), cycle day	Improved luteal phase recall; Reduced ovulation detection error by 2 days (vs. BBT) in individuals with high sleep timing variability [18]	Robust to sleep timing variations; Effective under free-living conditions [18]	Limited independent validation; Performance in irregular cycles not fully established [18]
Machine Learning (Random Forest) with Multi-Parameter Wearable Data [11]	Skin temperature, heart rate (HR), interbeat interval (IBI), electrodermal activity (EDA)	87% accuracy (3-phase classification); 68% accuracy (4-phase daily tracking) [11]	Multi-modal data fusion; Reduces self-reporting burden [11]	Accuracy drops with more granular (4-phase) classification [11]
Calendar-Based Methods (Rhythm Method) [84]	First day of last menstrual period, historical cycle length	N/A (Estimates fertile window only); Not suitable for irregular cycles [84]	Low cost, accessible; No special equipment needed [84]	Does not confirm ovulation; High error rate; Requires 6+ months of prior data [84]
Basal Body Temperature (BBT) [11]	Daily resting body temperature	Confirms ovulation post-occurrence (does not predict); Accuracy susceptible to sleep and environmental factors [11]	Long history of use; Confirms ovulation has occurred [11]	Does not predict ovulation; High measurement burden; Disrupted by sleep irregularities [18]
Natural Cycles App [84]	BBT, period data, optional LH tests	93% effective with typical use for pregnancy prevention [84]	FDA-cleared; Combines multiple data sources [84]	Primary focus is contraception; Requires consistent user input [84]

Detailed Experimental Protocols

For researchers aiming to implement or validate these methods in clinical or study settings, the following detailed protocols describe the standard procedures for key methodologies.

Protocol 1: Urinary Luteinizing Hormone (LH) Surge Detection

Objective: To precisely pinpoint the day of ovulation by detecting the luteinizing hormone (LH) surge in urine, which is considered a gold-standard biochemical marker for ovulation confirmation in research settings [5].

Materials:

• Urinary LH test kits (mid-stream or cassette)
• Timer
• Sample collection cups (if required by kit)
• Data recording form

Procedure:

• Initiation Timing: Begin testing daily on cycle day 10 or when cervical mucus becomes fertile (thin, clear, stretchy). Testing should be performed at approximately the same time each day, ideally in the afternoon [5].
• Sample Collection: Collect a urine sample in a clean, dry container. Avoid using first-morning urine as it may not detect the initial surge due to hormone concentration variations.
• Test Execution: Following the manufacturer's instructions, immerse the test strip in the urine sample or apply urine to the test cassette. Start the timer.
• Result Interpretation: Read the result at the time specified in the instructions (typically 5-10 minutes). A positive result, indicating the LH surge, is when the test line is as dark as or darker than the control line.
• Data Recording: Record the test result and corresponding cycle day. A positive test signifies that ovulation is likely to occur within the next 24-36 hours. The phase from 2 days before to 3 days after the positive test can be defined as the "ovulation phase" for research classification [11].

Protocol 2: Wearable-Based Phase Classification with Machine Learning

Objective: To classify menstrual cycle phases automatically using physiological data from a wrist-worn wearable device and a pre-trained machine learning model, minimizing participant burden and enabling tracking under free-living conditions [18] [11].

Materials:

• Research-grade wearable device (e.g., Empatica E4, Oura Ring, or EmbracePlus capable of measuring HR, IBI, skin temperature, EDA) [11]
• Data preprocessing and feature extraction pipeline
• Pre-trained Random Forest or XGBoost classification model [18] [11]
• Secure data storage and processing environment

Procedure:

• Data Collection:
  • Instruct participants to wear the device continuously, especially during sleep, for the duration of the study cycle(s).
  • Collect raw signals including heart rate, skin temperature, and electrodermal activity [11].

• Feature Extraction:

  • Fixed Window Technique: Segment data into non-overlapping windows corresponding to defined cycle phases (e.g., menstruation, follicular, ovulation, luteal). Calculate features like mean nocturnal heart rate, circadian rhythm nadir of heart rate (minHR), and skin temperature variability for each window [18] [11].
  • Sliding Window Technique: For daily phase prediction, use a sliding window (e.g., 7 days) to compute features continuously, allowing for dynamic phase transition detection [11].

• Model Application & Validation:

  • Input the extracted feature set into the pre-trained model (e.g., Random Forest for 3-phase classification: menstruation, ovulation, luteal).
  • Apply a leave-last-cycle-out or leave-one-subject-out cross-validation approach to evaluate model generalizability [11].
  • Output the predicted phase for each window or day alongside probability estimates.

• Performance Assessment:

  • Compare model predictions against a reference method (e.g., urinary LH test for ovulation). Calculate standard metrics including accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC) [11].

Protocol 3: Serum Progesterone Assay for Luteal Phase Confirmation

Objective: To confirm ovulation and a functional luteal phase by measuring mid-luteal phase serum progesterone levels, a direct hormonal validation method essential for high-quality research [5].

Materials:

• Phlebotomy kit
• Serum separation tubes
• Access to a clinical laboratory with a validated progesterone immunoassay
• Centrifuge
• Freezer for sample storage at -20°C if not analyzed immediately

Procedure:

• Timing of Sample Collection: Schedule the blood draw for the mid-luteal phase, approximately 7 days after a detected LH surge or predicted ovulation. In a 28-day cycle with ovulation on day 14, this would be around day 21 [5].
• Blood Collection: Draw a venous blood sample (e.g., 5-10 mL) into a serum separation tube using standard phlebotomy procedures.
• Sample Processing: Allow the blood to clot at room temperature for 15-30 minutes. Centrifuge the sample at a specified speed (e.g., 1300-2000 RCF) for 10 minutes to separate the serum.
• Assay Execution: Transfer the serum to a clean vial and analyze progesterone concentration using the laboratory's standard quantitative immunoassay (e.g., ELISA, CLIA). Strictly adhere to the manufacturer's protocol.
• Interpretation: A serum progesterone concentration typically exceeding 5-10 ng/mL (thresholds may vary by lab and assay) provides confirmation that ovulation has occurred and suggests a sufficient luteal phase [5]. This measurement serves as a ground truth for validating other phase-tracking methods.

Workflow Visualization

The following diagram illustrates the logical workflow for selecting an appropriate menstrual cycle tracking method based on research objectives and resources.

Method Selection Workflow

The Scientist's Toolkit: Essential Research Reagents & Materials

For researchers designing studies involving menstrual cycle phase determination, the following table details key reagents, materials, and technologies essential for implementing the protocols described.

Table 2: Essential Research Reagents and Materials for Menstrual Cycle Phase Determination

Item	Function / Application	Research Context & Considerations
Urinary LH Test Kits	Detects the luteinizing hormone (LH) surge to pinpoint ovulation [5].	Gold-standard for ovulation confirmation in non-clinical research. Cost-effective for longitudinal studies but requires daily participant compliance.
Enzyme-Linked Immunosorbent Assay (ELISA) Kits	Quantifies serum concentrations of progesterone and estradiol to confirm luteal phase functionality and hormonal status [5].	Provides direct hormonal measurement. Essential for establishing a eumenorrheic cycle in research; requires phlebotomy and laboratory facilities.
Research-Grade Wearable Sensors	Continuously collects physiological data (e.g., HR, HRV, skin temperature) under free-living conditions for input into ML models [18] [11].	Enables passive, long-term data collection. Key for developing and validating algorithmic approaches; device selection should match required signal types.
Basal Body Temperature (BBT) Thermometer	Measures subtle shifts in resting body temperature to retrospectively confirm ovulation [11].	Low-cost method for confirming ovulatory cycles. Subject to measurement noise from sleep disruptions; less reliable for prediction [18].
Random Forest / XGBoost Classifiers	Machine learning algorithms that integrate physiological features to classify menstrual cycle phases [18] [11].	Core of modern algorithmic tracking. Requires expertise in feature engineering and model validation (e.g., leave-one-subject-out cross-validation).

Reporting Standards and Transparent Communication of Limitations

In the rapidly advancing field of menstrual cycle research, the accurate characterization and reporting of cycle phases represents a fundamental methodological challenge with significant scientific and clinical implications. Recent systematic evaluations have revealed a concerning trend: the widespread use of assumed or estimated menstrual cycle phases to characterize ovarian hormone profiles, an approach that amounts to little more than educated guessing [5]. This practice persists despite clear evidence that calendar-based counting methods alone demonstrate less than 30% accuracy in predicting actual ovulation when verified against hormonal biomarkers [85]. The consequences of these methodological shortcomings extend beyond theoretical concerns, potentially compromising female athlete health, training recommendations, performance optimization, and injury prevention strategies [5].

The problem is further exacerbated by substantial inconsistencies in how laboratories operationalize and report menstrual cycle phases across studies [6] [12]. Without standardized protocols and transparent reporting of methodological limitations, the field experiences significant confusion in the literature and frustrated attempts at systematic reviews and meta-analyses [12]. This article establishes comprehensive application notes and protocols designed to address these critical gaps by providing researchers with standardized tools for enhancing methodological rigor, particularly focusing on the transparent communication of limitations when ideal measurement standards cannot be fully implemented.

Quantitative Assessment of Current Methodological Approaches

Accuracy of Phase Determination Methods

Table 1: Comparative Accuracy of Menstrual Cycle Phase Determination Methods

Method Category	Specific Method	Reported Accuracy	Key Limitations	Appropriate Use Cases
Counting Methods	Forward-counting	<30% (vs. LH test) [85]	High cycle variability affects precision	Initial screening only
	Backward-counting	<30% (vs. LH test) [85]	Requires predictable luteal phase	Initial screening only
Hormone Monitoring	Urine LH testing	>95% (with protocol) [85]	Cost, participant burden	Ovulation confirmation
	Quantitative urine hormones (Mira)	Under validation [13]	Limited published validation	Research settings
Physiological Tracking	BBT	76.92-99% [11]	Confirms ovulation post-hoc	Cycle pattern identification
	Wearable sensors (machine learning)	68-87% [11]	Emerging validation	Ambulatory monitoring

Methodological Validation Metrics

Table 2: Validation Metrics for Emerging Monitoring Technologies

Technology	Validation Reference	Sample Size	Accuracy	Specificity/Sensitivity	Limitations
Wrist-worn device (RF model)	Fixed window (3 phases) [11]	65 cycles/18 subjects	87%	AUC-ROC: 0.96	Limited sample size
Wrist-worn device (RF model)	Sliding window (4 phases) [11]	65 cycles/18 subjects	68%	AUC-ROC: 0.77	Reduced granularity accuracy
Mira Monitor (Quantitative urine)	Ultrasound validation protocol [13]	Target: 150 cycles	Under study	Correlation with serum/ultrasound	Preliminary results pending

Experimental Protocols for Menstrual Cycle Research

Gold Standard Validation Protocol

The Quantum Menstrual Health Monitoring Study establishes a comprehensive protocol for validating quantitative hormone monitoring against gold standard references [13]. This approach characterizes patterns in urine hormones (FSH, E13G, LH, PDG) that predict and confirm ovulation, referenced to both serum hormones and ultrasound-confirmed ovulation day.

Participant Recruitment and Group Stratification:

• Group 1: Regular cycles (24-38 days) for establishing baseline reference values
• Group 2: Polycystic ovarian syndrome (PCOS) with irregular cycles
• Group 3: Athletes with irregular cycles associated with high exercise levels

Longitudinal Monitoring Protocol:

• Duration: 3 consecutive menstrual cycles
• Urine Hormone Monitoring: Daily collection with Mira monitor
• Ultrasound Confirmation: Serial follicular tracking ultrasounds in community clinic
• Serum Correlation: Periodic blood draws for hormone correlation
• Supplementary Data: Bleeding patterns tracked via validated Mansfield-Voda-Jorgensen Menstrual Bleeding Scale [13]

Validation Metrics:

• Primary outcome: Correlation between urine hormone patterns and ultrasound day of ovulation
• Secondary outcome: Agreement between urine and serum hormone concentrations
• Statistical power: 150 cycles provide 80% power to detect 0.5-day differences in ovulation timing

Integrated Hormone and Symptom Monitoring Framework

Figure 1: Gold Standard Validation Workflow for Menstrual Cycle Monitoring

Machine Learning Protocol for Phase Classification

Emerging approaches apply machine learning to classify menstrual cycle phases using physiological signals from wearable devices [11]. This protocol enables automated phase tracking while reducing participant burden.

Data Collection Specifications:

• Devices: Wrist-worn sensors (E4, EmbracePlus)
• Signals: Heart rate (HR), interbeat interval (IBI), electrodermal activity (EDA), skin temperature, accelerometry (ACC)
• Duration: 2-5 months of continuous monitoring
• Reference Standard: Urinary luteinizing hormone (LH) tests for ovulation confirmation

Feature Engineering and Model Training:

• Fixed Window Approach: Features extracted from non-overlapping cycle phases
• Rolling Window Approach: Sliding window for daily phase tracking
• Validation: Leave-last-cycle-out and leave-one-subject-out approaches
• Algorithms: Random forest, logistic regression, other classifiers compared

Performance Metrics:

• Three-phase classification (period, ovulation, luteal): 87% accuracy
• Four-phase classification (period, follicular, ovulation, luteal): 68% accuracy
• Area under ROC curve: 0.96 for three-phase classification

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Essential Research Materials for Menstrual Cycle Study Protocols

Category	Specific Tool/Reagent	Research Application	Key Considerations
Hormone Validation	Urine LH test kits	Ovulation confirmation	Quality affects accuracy; store properly
	Quantitative hormone monitor (Mira)	Daily hormone patterns	Multiple hormone measurement; cost factor
	Serum hormone assays	Gold standard reference	Requires venipuncture; laboratory processing
Physiological Monitoring	Wrist-worn sensors (E4, EmbracePlus)	Continuous physiological data	Multi-parameter signals; participant compliance
	Basal body temperature devices	Temperature shift detection	Measurement consistency critical
	OvuSense vaginal sensor	Core temperature monitoring	99% ovulation detection accuracy [11]
Cycle Tracking	Validated bleeding scales (Mansfield-Voda-Jorgensen)	Menstrual bleeding quantification	Standardized assessment
	Daily symptom tracking apps	Prospective symptom monitoring	Avoids recall bias; essential for PMDD diagnosis [6]
Data Analysis	Carolina Premenstrual Assessment Scoring System (C-PASS)	PMDD/PME diagnosis	Standardized scoring system available [6]
	Random forest classifiers	Machine learning phase detection	Handles multi-parameter physiological data [11]

Standardized Reporting Framework for Methodological Limitations

Figure 2: Limitations Reporting and Communication Framework

Mandatory Reporting Elements

Regardless of the methodological approach employed, researchers must transparently report specific elements to enable proper interpretation of findings and facilitate meta-analyses.

For All Studies:

• Cycle Phase Determination Method: Explicit description of criteria used for phase classification
• Verification Procedures: Detail any hormonal or physiological verification methods employed
• Participant Cycle Characteristics: Report cycle regularity, length variability, and exclusion criteria
• Terminology Precision: Use "naturally menstruating" for calendar-based counting versus "eumenorrheic" for hormonally confirmed cycles [5]

When Using Indirect Methods:

• Accuracy Limitations: Explicitly acknowledge the <30% accuracy of counting methods in predicting actual fertility [85]
• Potential Misclassification: Quantify potential impact of phase misclassification on primary outcomes
• Clinical Significance: Discuss how methodological limitations affect clinical applicability

Statistical Considerations for Transparent Reporting

Appropriate statistical approaches must align with the menstrual cycle as a within-person process [6]. Between-subject designs conflate within-subject variance (changing hormone levels) with between-subject variance (individual baseline symptoms), fundamentally compromising validity.

Minimum Standards:

• Repeated Measures: Three observations per person minimum to estimate random effects
• Multi-Cycle Designs: Three or more observations across two cycles for reliable between-person difference estimation
• Person-Centered Analysis: Graph individual patterns before group-level analysis to detect outliers and relevant patterns
• Cycle Day Coding: Combined forward-count (days 1-10) and backward-count (final 10 days) methods for optimal phase alignment [6]

The establishment and adherence to standardized reporting protocols represents an essential step forward for menstrual cycle research. By implementing these application notes and protocols, researchers can significantly enhance the methodological rigor of their studies while enabling proper interpretation of findings within methodological constraints. The transparent communication of limitations is not an admission of methodological weakness but rather a commitment to scientific integrity and cumulative knowledge advancement.

As the field continues to evolve with emerging technologies such as quantitative hormone monitors and machine learning approaches, these reporting standards provide a framework for validating new methods against established references. Through consistent application of these protocols across laboratories and research groups, the field can overcome current limitations in comparability and reproducibility, ultimately accelerating our understanding of menstrual cycle effects on health, performance, and disease.

Conclusion

Accurate computational determination of menstrual cycle phases is no longer a convenience but a necessity for rigorous biomedical research and drug development. This synthesis demonstrates that while machine learning models leveraging multi-modal physiological data can achieve high accuracy—exceeding 87% in some studies for phase classification—their success is contingent on moving beyond simplistic calendar methods and embracing robust validation. Future directions must focus on developing more adaptive models for individuals with irregular cycles, integrating novel contactless biosensing technologies, and standardizing validation protocols across the field. By adopting these sophisticated computational protocols, researchers can generate higher-quality, more reliable data, ultimately accelerating innovation in women's health and ensuring that female biology is accurately represented in clinical science.

References

• 1. Validity of Self-Reported Menstrual Cycle Length [https://www.sciencedirect.com/science/article/abs/pii/S1047279706001335]
• 2. Accuracy of Reporting of Menstrual Cycle Length - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC3693484/]
• 3. Using Hormone Data and Age to Pinpoint Cycle Day within ... [https://www.mdpi.com/1648-9144/59/7/1348]
• 4. Accuracy of Calendar-Based Methods for Assigning ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC3658377/]
• 5. Why We Must Stop Assuming and Estimating Menstrual ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12152053/]
• 6. How to study the menstrual cycle - PubMed Central - NIH [https://pmc.ncbi.nlm.nih.gov/articles/PMC8363181/]
• 7. The potential bidirectional relationship between long ... [https://www.nature.com/articles/s41467-025-62965-7]
• 8. A survey assessing the impact of symptoms related to ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12398178/]
• 9. Prevalence of Hypothyroidism in Women Presenting with ... [https://healthcare-bulletin.co.uk/article/prevalence-of-hypothyroidism-in-women-presenting-with-menstrual-irregularities-4060/]
• 10. Determining Menstrual Cycle Phase - PubMed Central - NIH [https://pmc.ncbi.nlm.nih.gov/articles/PMC10714354/]
• 11. Machine learning-based menstrual phase identification ... [https://www.nature.com/articles/s44294-025-00078-8]
• 12. How to study the menstrual cycle: Practical tools and ... [https://www.sciencedirect.com/science/article/abs/pii/S0306453020303188]
• 13. Establishing a Gold Standard for Quantitative Menstrual ... [https://www.mdpi.com/1648-9144/59/9/1513]
• 14. Advancements in biosensing of hormones [https://www.sciencedirect.com/science/article/abs/pii/S0165993624000827]
• 15. Cycle Education Flow Chart [https://www.rhmgyn.com/flowchart]
• 16. Survey Analysis of Quantitative and Qualitative Menstrual ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10534579/]
• 17. Oura Ring as a Tool for Ovulation Detection [https://www.jmir.org/2025/1/e60667/]
• 18. Machine learning model for menstrual cycle phase ... [https://www.sciencedirect.com/science/article/pii/S0010482525000551]
• 19. A comparison of methods to predict ovulation day ... [https://pubmed.ncbi.nlm.nih.gov/40668218/]
• 20. Innovative Approaches to Menstruation and Fertility Tracking Using... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10905339/]
• 21. Journal of Medical Internet Research - Wearable Reveal... Sensors [https://www.jmir.org/2019/4/e13404]
• 22. Advances in biomonitoring technologies for women's health [https://pmc.ncbi.nlm.nih.gov/articles/PMC12475061/]
• 23. Adaptive Edge-Federated AI Framework for Contactless ... [https://www.sciencedirect.com/science/article/pii/S2215016125005096]
• 24. Pulse Rate Measurement During Sleep Using Wearable , and... Sensors [https://www.nature.com/articles/s41598-017-01433-9?error=cookies_not_supported]
• 25. Smart Rings & Watches for Cycle Tracking 2025 [https://www.allthingsdaysie.com/en/post/smart-rings-und-watches-for-cycle-tracking-2025-a-complete-guide]
• 26. Effect of Different Phases of Menstrual Cycle on Heart Rate ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4625231/]
• 27. Nocturnal finger skin temperature in menstrual cycle tracking [https://bmcwomenshealth.biomedcentral.com/articles/10.1186/s12905-019-0844-9]
• 28. The effect of menstrual on maximal breath-hold time cycle [https://pubmed.ncbi.nlm.nih.gov/31923591/]
• 29. Heart Rate Variability Measurements Across the Menstrual ... [https://www.mdpi.com/2075-4663/13/6/185]
• 30. Using wearable skin temperature data to advance tracking ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11294004/]
• 31. Journal of Medical Internet Research - Innovative Approaches to... [https://www.jmir.org/2024/1/e45139/]
• 32. mcPHASES: A Dataset of Physiological , Hormonal, and Self-reported... [https://physionet.org/content/mcphases/1.0.0/]
• 33. Machine learning model for menstrual cycle phase ... [https://pubmed.ncbi.nlm.nih.gov/39889448/]
• 34. "Ovulation prediction using random forest algorithm on ... [https://digitalcommons.unf.edu/etd/1229/]
• 35. Machine learning-based evaluation of application value ... [https://www.sciencedirect.com/science/article/pii/S1876382023000872]
• 36. the optimal protocol choice for IVF-ET [https://www.sciencedirect.com/science/article/pii/S2090123225008355]
• 37. Fixed vs Rolling Time Windows? - Issue 175 [https://dataanalysis.substack.com/p/fixed-vs-rolling-time-windows-issue]
• 38. What is a rolling window in time series analysis? [https://milvus.io/ai-quick-reference/what-is-a-rolling-window-in-time-series-analysis]
• 39. Creating Rolling Window Features for Time Series [https://businessanalyticsinstitute.com/rolling-window-features-time-series/]
• 40. Rolling Window Analysis [https://questdb.com/glossary/rolling-window-analysis/]
• 41. Introduction to feature engineering for time series forecasting [https://medium.com/data-science-at-microsoft/introduction-to-feature-engineering-for-time-series-forecasting-620aa55fcab0]
• 42. Analyzing physiological signals recorded with a wearable ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10621043/]
• 43. Feature Engineering Techniques For Time Series Data [https://www.analyticsvidhya.com/blog/2019/12/6-powerful-feature-engineering-techniques-time-series/]
• 44. Practical Guide for Feature Engineering of Time Series Data [https://dotdata.com/blog/practical-guide-for-feature-engineering-of-time-series-data/]
• 45. A critical signal for phenotype transition driven by negative ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10788427/]
• 46. Why Individualized Care Is the Future of Women's Health [https://activated.health/why-individualized-care-is-the-future-of-womens-health/]
• 47. Anovulatory Bleeding - StatPearls - NCBI Bookshelf [https://www.ncbi.nlm.nih.gov/books/NBK549773/]
• 48. Day-to-day variability in sleep parameters and depression ... [https://www.nature.com/articles/s41746-021-00400-z]
• 49. Mapping the Neural Basis of Wake Onset Regularity and Its ... [https://www.mdpi.com/2624-5175/7/1/15]
• 50. Sleep and Circadian Variability in People with Delayed ... - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC5514558/]
• 51. Monitoring Sleep Quality Through Low α-Band Activity in ... [https://www.jmir.org/2025/1/e67188/]
• 52. Patterns of missing data in the use of the endometriosis ... [https://bmcwomenshealth.biomedcentral.com/articles/10.1186/s12905-018-0578-0]
• 53. Real-world menstrual characteristics of more than 600,000... cycle [https://www.nature.com/articles/s41746-019-0152-7?error=cookies_not_supported&code=e3833bca-e4e5-4efc-85df-384944c5f240]
• 54. Realignment and multiple imputation of longitudinal data [https://pmc.ncbi.nlm.nih.gov/articles/PMC4059763/]
• 55. mcPHASES: A Dataset of Physiological, Hormonal, and ... [https://www.physionet.org/content/mcphases/]
• 56. Understanding the Phases of the Menstrual Cycle [https://helloclue.com/articles/cycle-a-z/the-menstrual-cycle-more-than-just-the-period]
• 57. Modelling menstrual cycle length in athletes using state- ... [https://www.nature.com/articles/s41598-021-95960-1]
• 58. UEFA consensus statement on menstrual cycle tracking in ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12458775/]
• 59. Implementing menstrual cycle tracking: A pilot concept ... [https://www.sciencedirect.com/science/article/pii/S1440244024001154]
• 60. Predicting polycystic ovary syndrome with machine ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10866556/]
• 61. A machine learning approach for non-invasive PCOS ... [https://www.nature.com/articles/s41598-025-10453-9]
• 62. How NC° Perimenopause works – Customer Support [https://help.naturalcycles.com/hc/en-us/articles/30400959172893-How-NC-Perimenopause-works]
• 63. Case Reports from Women Using a Quantitative Hormone ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10608103/]
• 64. Menstrual tracking app data is a 'gold mine' for advertisers ... [https://pharmacyupdate.online/2025/06/menstrual-tracking-app-data-is-a-gold-mine-for-advertisers-that-risks-womens-safety-report/]
• 65. All Eyes on my Period? Period tracking apps and the future ... [http://privacyinternational.org/long-read/5593/all-eyes-my-period-period-tracking-apps-and-future-privacy-post-roe-world]
• 66. Intimate harms and menstrual cycle tracking apps [https://www.sciencedirect.com/science/article/pii/S0267364924001043]
• 67. Federated learning for predicting clinical outcomes in ... [https://www.nature.com/articles/s41591-021-01506-3]
• 68. Adaptive Edge-Federated AI Framework for Contactless ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12593638/]
• 69. Federated Learning for privacy-Friendly Health Apps [https://www.mdpi.com/2224-2708/14/1/11]
• 70. "Menstrual cycle prediction from physiological data using ... [https://digitalcommons.unf.edu/etd/1242/]
• 71. A Federated Learning Emulation and Evaluation Testbed ... [https://arxiv.org/html/2509.00621v1]
• 72. Mathematical Modeling and Simulation Provides Evidence ... [https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2021.613048/full]
• 73. Complexities in salivary and urinary methodology for ... [https://gremjournal.com/journal/01-2025/complexities-in-salivary-and-urinary-methodology-for-ovulation-and-menstrual-cycle-hormone-detection-a-scoping-review/]
• 74. Validating At-Home Urinary Hormone Measurements in ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12040536/]
• 75. Quantitative detection of pre-ovulatory luteinizing hormone ... [https://www.sciencedirect.com/science/article/pii/S0039914024009469]
• 76. A Clinical Assessment of the Outcomes of Intrauterine ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12158816/]
• 77. Choosing the Right Metrics: Recall, Precision, PR Curve ... [https://medium.com/@juanc.olamendy/choosing-the-right-metrics-recall-precision-pr-curve-and-roc-curve-explained-682259961cbe]
• 78. Decoding female physiology: a scoping review on the use ... [https://www.medrxiv.org/content/10.1101/2025.09.24.25336575v2.full-text]
• 79. The impact of menstrual cycle phase and symptoms on ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12500648/]
• 80. Power in the flow: how menstrual experiences shape ... [https://www.frontiersin.org/journals/sports-and-active-living/articles/10.3389/fspor.2025.1519825/full]
• 81. Can the Menstrual Cycle Phase Impact Sports Performance? [https://blogs.bmj.com/bjsm/2025/11/03/can-the-menstrual-cycle-phase-impact-sports-performance/]
• 82. Making experimental data tables in the life sciences more FAIR [https://pmc.ncbi.nlm.nih.gov/articles/PMC7736789/]
• 83. Comparing experimental conditions using modern statistics [https://link.springer.com/article/10.3758/s13428-020-01471-8]
• 84. Calendar Method: Birth Control, Calendar & Effectiveness [https://www.naturalcycles.com/cyclematters/calendar-method]
• 85. Standardized protocols for characterizing women's fertility [https://www.sciencedirect.com/science/article/pii/S0018506X15300763]