Analyzing the Atypical – Methods for studying the menstrual cycle in adolescents

Abstract

Background.

The female pubertal transition is characterized by a rapidly changing hormone milieu, which is heavily influenced by the first menstrual cycle – menarche. The first year following menarche is associated with menstrual cycles that are irregular and anovulatory. Peripuberty also marks the beginning of a female-biased risk for suicidality and depression, suggesting some influence by the menstrual cycle and ovarian hormone fluctuations. However, there are limited methods and guidelines for studying the menstrual cycle and related affective symptoms in this developmental window. Thus, this study’s objective was to identify the most accurate methods for detecting ovulation in irregular cycles (Part 1) and develop guidelines based on these methods for determining menstrual cycle phases. These methods were applied to investigate hormones and affective symptoms based on cycle phase and ovulation status in a sample of peripubertal females (Part 2).

Methods.

Thirty-two peripubertal females (ages 11-14) provided daily urine samples of estrogen (E1G) and progesterone (PdG) metabolites and luteinizing hormone (LH), and ratings of affective symptoms for one menstrual cycle. Ten literature-derived methods for determining the presence of an LH-peak or PdG rise were compared, focusing on their feasibility for psychological research.

Results.

Methods by Sun et al. (2019) and Park et al. (2007) most accurately detected PdG rises and LH peaks in this sample, identifying 40.6% of cycles as ovulatory. As expected, ovulatory participants showed greater LH in the periovulatory phase (p=.001), greater PdG in the mid-luteal phase (p<.0001), and greater E1G in the periovulatory phase (p=.001) compared with anovulatory participants. Exemplary methods to compare psychological symptoms between both groups are provided.

Conclusions.

Recommendations and guidelines for studying the menstrual cycle in irregular cycling adolescents are offered. Novel methods for ovulation detection identified phase-specific hormonal patterns in anovulatory and ovulatory adolescent cycles.

1. INTRODUCTION

The pubertal transition (peripuberty) is characterized by extensive physical maturation and a rapidly changing reproductive hormone milieu. Aside from these biological changes, peripuberty marks the beginning of a sex disparity in risk for psychopathology that continues throughout the female reproductive lifespan. Rising ovarian hormone levels and their increased fluctuation across the menstrual cycle may contribute to this abrupt rise in vulnerability to psychopathology (i.e., suicidal attempts, depression) in peripubertal female adolescents (Angold and Costello, 2006; Owens et al., 2020; Schiller et al., 2016; Thapar et al., 2012). Given the challenges of assessing the often anovulatory and irregular cycles of peripubertal participants, psychological symptoms accompanying hormone changes are rarely studied. The present study sought to fill this critical research gap by providing novel recommendations for studying the menstrual cycle in peripubertal participants, in which irregular and anovulatory cycles are common.

1.1. The peripubertal menstrual cycle

The onset of menstrual cycling, or menarche, is a biological, and for some, also a psychosocial milestone in female adolescent development that typically occurs between the ages of 12 and 13 (American Academy of Pediatrics et al., 2006). Across the menstrual cycle, hormones, including estrogens, progesterone, luteinizing hormone (LH) and follicle stimulating hormone (FSH), finely coordinate the growth of follicles and the release of an oocyte into the uterus (ovulation) for fertilization (Bale and Epperson, 2017). Absent a pregnancy, the endometrium is excreted through menstrual bleeding and a new cycle starts. These hormones fluctuate systematically throughout the cycle and thereby characterize distinct phases of the cycle (mid-follicular, periovulatory, mid-luteal, and perimenstrual phases).

The first-year post-menarche, also referred to as the first gynecological year, is associated with irregular cycle lengths and frequent anovulatory cycles (Gunn et al., 2018; WHO, 1986). The mean cycle length in the first gynecological year ranges from 32 to 61 days with high variability (Gunn 2018). This contrasts with adult (>18 years) menstrual cycles that have a more stable mean length of approximately 28 days, with normal variations between 22 and 36 days (Fehring et al., 2006). Gunn et al. (2018) also reported that prevalence of ovulatory cycles in the first gynecological year ranges from 0 to 45%. Furthermore, shorter luteal phases and/or lower progesterone concentrations compared to adults are common (Carlson and Shaw, 2019). In general, female adolescents show lower mean output of estradiol and progesterone concentrations than adults throughout the cycle, yet equivalent concentrations of LH across the cycle (Carlson and Shaw, 2019; Sun et al., 2019a). Cycle irregularity and anovulation often continue after the first gynecological year. It can take six years until normal cycle length is established (American Academy of Pediatrics et al., 2006). Anovulatory cycles show different hormonal patterns from ovulatory cycles; specifically, lower overall concentrations and smaller amplitudes of estradiol, and LH and almost no progesterone output (Hambridge et al., 2013).

Characterizing the menstrual cycle and distinguishing its phases is critical to investigating underlying hormonal patterns and related risk for psychological symptom changes during peripuberty (see 1.2). Recently, guidelines on how to study the menstrual cycle in adults have been developed (Schmalenberger et al., 2021), and thresholds for biological markers to determine cycle phases have been proposed for various specimens (i.e., blood, serum, and urine) (Arslan et al., 2022; Barbieri, 2014; Fiers et al., 2017) using different analysis methods (Tivis et al., 2005). Yet, these recommendations are unsuitable for peripubertal cycles due to their irregularities in ovulation, cycle length, and hormone concentrations discussed previously (Yu et al., 2017). Frequent transvaginal ultrasounds are considered the gold standard for determining ovulation (Ecochard et al., 2001; Lynch et al., 2014); however, this approach is not feasible for most research studies. Alternative methods for determining ovulation in the early years post-menarche are scarce and require additional study guidelines.

1.2. Impact of the menstrual cycle characteristics on psychological symptoms

Across the last quarter century, a series of experimental studies have demonstrated that affective symptoms tied to reproductive events are caused by an aberrant “hormone sensitivity” to normal hormone changes, although the underlying neurobiology of this sensitivity remains poorly understood. Hormone sensitivity has been experimentally demonstrated in patients with premenstrual dysphoric disorder (PMDD; luteal phase confinement of affective symptoms; Schmidt et al. (2017)), perinatal depression (Bloch, 2000; Schiller et al., 2022), and perimenopausal-onset depression (Schmidt et al., 2015). In vulnerable females, hormone-induced affective symptoms can be triggered by periovulatory surges in progesterone metabolites (Martinez et al., 2016; Schiller et al., 2014; Schmidt et al., 1991) as well as by perimenstrual estradiol withdrawal (Barone et al., 2022; Eisenlohr-Moul et al., 2022). In addition to rising hormone concentrations and variability that accompany puberty, the cyclical sources of hormone-sensitive affective symptoms described above are expected to emerge and increase risk of psychopathology at peripuberty.

We recently demonstrated that many peripubertal females (pre- and post-menarche) show affective sensitivity to weekly changes in estrone and testosterone. However, this investigation was restricted to weekly salivary hormone and affective symptom assessments and did not consider menstrual cycle characteristics (Andersen et al., 2022). Few studies have examined psychological symptoms accompanying hormone change in peripubertal females, possibly due to challenges in detecting ovulation and cycle irregularity in the first-year post-menarche. While multiple ovulation detection methods have been used in adult research, the results vary substantially and have not been tested in an adolescent population (Lynch et al., 2014).

1.3. Research objectives of the present study

The absence of guidelines for investigating the menstrual cycle in adolescents (including ovulation detection) inhibits the ability to elucidate the effects of the menstrual cycle and associated hormonal fluctuation in the emergence of adolescent psychopathology. The objectives of the present study were to 1) review methods for ovulation detection and empirically determine the most reliable ovulation detection method to develop guidelines and recommendations for determining ovulation and to 2) adapt guidelines for determining menstrual cycle phases for adolescents or adults with frequent anovulatory cycles, which hormone patterns that do not map onto the typical cycle. As an exploratory example, we applied these recommendations to a sample of peripubertal participants and investigated effects of ovulatory status on psychological symptoms typically associated with the menstrual cycle. Daily dried urine samples of ovulatory hormones and menstrual-related psychological symptom assessments in a sample of peripubertal girls (≤ 1-year post menarche) were analyzed to identify the most accurate method for detecting ovulation status (Part 1) and to characterize menstrual cycle phases (Part 2), and to exemplary compare symptom patterns between ovulatory and anovulatory participants.

2. METHODS

2.1. Study design and procedure

Following a screening questionnaire and enrollment session, eligible participants completed daily menstrual-related psychological symptom ratings and hormone measurements for 28 days or one menstrual cycle (up to 48 days). Forty-nine participants answered daily questionnaires on current symptoms for 28 days or one menstrual cycle on a study iPad. After completion and applying additional exclusion criteria (see section 2.2.), 32 participants provided sufficient cycle data to be included in this study. The first 24 participants (n = 15 of final sample) provided data for 28 days; however, because several participants showed substantially longer cycle lengths (more than 41 days), this assessment time frame was modified to assess a full menstrual cycle of the individual up to 48 days. The protocol was approved by the Institutional Review Board at University of North Carolina at Chapel Hill and was conducted in accordance with the Declaration of Helsinki.

2.2. Participant recruitment and inclusion

Peripubertal participants between the ages of 11 and 14 were recruited through flyers posted in schools and the local community, mass emails to university members, postings on middle school parent websites, and word-of-mouth. Participants were assigned female at birth, physically healthy, and undergoing a natural pubertal transition. Participants were mid-puberty, determined by pubic hair growth and breast development underway but not complete according to the pubertal development scale (Petersen et al., 1988) and Tanner staging with line drawings, which was self and parent-reported. We excluded participants who were over 15 months post-menarche (based on parent’s report) at the enrollment session, non-English speaking, pregnant, taking hormonal contraception, taking pharmaceutical regimens that directly alter cardiovascular or neurological function, a reported personal history of any chronic medical condition, psychotic or bipolar disorders or active suicidality (assessed using an abbreviated Structured Clinical Interview for DSM-V (SCID-CV) (First et al., 2017) and Columbia-Suicide Severity rating scale (Posner et al., 2008)). Parents provided written consent and adolescent participants provided written assent to participate in the study. Participants were compensated with a $180.00 gift card for full compliance.

2.3. Biological measures

To examine ovarian hormone concentrations and determine ovulation status for cycle phasing, urinary metabolites of estradiol (estrone-3-glucuronide, E1G) and progesterone (pregnanediol glucuronide, PdG), luteinizing hormone (LH) and creatinine were measured daily. Urine samples were collected on filter paper immediately upon awakening to capture the peak hormone response and dried completely (at least 24 hours) before being stored in the participant’s home freezer. Pregnanediol (PdG), the primary urinary metabolite of progesterone, is particularly sensitive for detecting ovulation (Carlson and Shaw, 2019). Further, dried urine sampling offers a noninvasive, non-burdensome method with superior feasibility for daily collections. Precision is comparable to serum concentrations (Newman et al., 2019) and liquid urine samples (unpublished data provided by ZRT laboratory; Appendix C), while offering superior stability. Please refer to Appendix C for validation studies provided by ZRT laboratory comparing dried urine and liquid urine analyses of E1G, PdG and LH for raw and creatinine corrected values. At the end of the collection period, samples were retrieved and transferred to a −80°C laboratory freezer before being sent to ZRT laboratory (Beaverton, OR) for analysis.

A 6-mm hole punch, 96 well-fritted filter block and a buffered extraction solution were used to prepare the samples for analysis. Enzyme-linked immunosorbent (ELISA) assays were used for E1G, PdG and LH and a colorimetric assay for creatinine was used. Intra-assay and inter-assay coefficients of variation for EIG, PdG, and LH were between 0.6-11.6% and 4.9-18.7%, respectively. Limits of detection were 1.07 mIU/mL for LH, 4.02 ng/mL for E1G, 57.58 ng/mL for PdG, and 0.026 mg/mL for creatinine. Hormone measurements were corrected for creatinine using the following equation: analyte divided by creatinine after matching volume units.

2.4. Assessment of DRSP symptoms

Daily Record of Severity of Problems (DRSP):

10 items modified from the DRSP were used to assess menstrual-related psychological symptoms (DRSP symptoms; Endicott et al., 2006). The selected 10 items included a variety of symptoms typically associated with premenstrual dysphoric disorder, including depression, hopelessness, anxiety, mood lability, anger, low interest, sensitivity to rejection, interpersonal conflict, feeling overwhelmed, and difficulty concentrating. Participants rated symptoms using a 6-point scale ranging from 1 (not at all) to 6 (extreme). Test-retest reliability for the DRSP is above 0.70, internal consistency scores are high, and correlations with measures such as the Hamilton Depression Rating Scale or the Quality of Life Enjoyment and Satisfaction Questionnaire are moderate to high (Endicott et al., 2006).

2.5. Part 1: Determining ovulation status

2.5.1. Protocol for identifying ovulation

Ovulation status can be determined from ovulatory hormone concentrations if the following occur: (1) an LH peak that is (2) closely followed by a rise in progesterone. Available methods for detecting ovulation status are either based on adult hormone level thresholds or yield no consistent procedure (Lynch et al., 2014). Thus, we identified and systematically compared ten¹ determination methods in the literature to isolate the most accurate methods for identifying an LH peak and PdG rise (Apter et al., 1978; Brown, 1977; Hoff et al., 1983; Johansson et al., 1971; Park et al., 2007; Sun et al., 2019b; Testart et al., 1981; Zhang et al., 2008). Descriptions of the determination methods and their comparison can be found in Table 1.

Table 1.

Overview of literature-derived methods to determine LH peak and PDG rise

General information on method			Application to adolescent sample
Reference	Detection algorithm	Adolescent specifics	% of participants with detected peak	Mean number of LH peaks	Points from rating
LH peak detection
Park et al. (2007)	LH-surge defined as 2.5 increase in LH from that of mean of five preceding days	No
			84%	1.97	42
Brown (1977) (as cited by Lynch)	LH surge defined as first LH value exceeding +3 SD above averages of both the immediately preceding and following five days that is also of ≤ 2 days duration	No	41%	0.5	34
Johansson et al. (1971)	LH-surge defined as 4-fold increase in LH from the day before	No	62%	1.38	24
Apter et al. (1978)	Highest value	Yes	100%	1	-
Hoff et al. (1983)	First LH value exceeding the mean + 2 sd of the six preceding values	No	94%	3.19	-
Testart et al. (1981) (as cited by Lynch)	LH surge defined as LH level > 180% above the average of the preceding four LH values	No	94%	3.53	-
Zhang et al (2008)	≥ mean + 3 SD of the five preceding values	Yes	94%	3.2	-
Progesterone rise detection
			% of participants with detected rise	% of PdG rises, that fall in last 14 days of cycle	Points from rating
Sun et al. (2019)	Threefold increase in urine Pd above mean follicular phase (FP) levels (modified Kassam algorithm)	Yes	50%	90%	31
Zhang et al (2008)	Lowest 5-day average as baseline; rise is threefold increase for at least three days, and only for PdG: ≥ 1.0 mg/g Cr	Yes	34%	85%	20
Kassam et al. (1996)	Threefold increase for three or more days compared of average of day 6-10	No	25%	100%	15

Note.

^*methods were applied to the dataset and compared with the rating system described in the manuscript. Methods are selected from comparison paper of ovulation detection algorithms (Lynch et al., 2014), review on peripubertal transition and hormones (Carlson and Shaw, 2019) and snowball search for further methods.

(1). LH peak detection protocol:

For normally cycling adults, LH concentrations range from 18-25 mIU/mL and triple at peak to approximately 60 mUI/mL (Barbieri, 2014). Because LH levels are likely more variable in adolescents (unpublished data), a relational criterion dependent on individual mean LH concentrations is necessary, rather than a threshold based on absolute concentrations. We applied seven relational methods used in previous publications to identify the most accurate method for the present sample¹.

(2). PdG rise in luteal phase:

Similar to LH, PdG concentrations are lower in peripuberty compared to adulthood. Therefore, absolute threshold measures are not reliable, and a relational method is necessary for determining the rise of PdG during the luteal phase.

Criteria for evaluating these methods were based on 1) whether the prevalence of detected LH peaks is plausible for this age group (less than 80% of participants showing a peak), 2) a distinct LH peak (meaning only peak across the cycle) was isolated without detecting multiple minor rises, and 3) the majority of the PdG rise occurred during the luteal phase. After discarding methods that did not fulfill these criteria, a total of 6 methods were systematically tested against each other (marked in table 1). Two experts (HK and KB) independently applied each method to each participant of our sample and used the following rating system to assess how precisely each method detected an LH peak and PdG rise (refer to Figure 1 for an example). Two points were given when the method accurately detected an LH peak (or PdG rise), one point was given if the method identified a visible peak/rise but not exclusively, and no points were issued if the method failed to detect or incorrectly detected a peak/rise. A third researcher (EA) resolved rating disagreements. The methods with the highest score for accurately detecting an LH peak and PdG rise were used for all further analyses in this paper.

Figure 1. Rating System for determining method accuracy.

LH concentration curve (mIU/mL, creatinine corrected) of an example participant with peaks detected by different methods in different colors. The x-axis reflects the backward-count cycle days leading up to the subsequent onset of menses (day 0). Refer to the text for description of the rating system.

(3). Combined ovulation status:

The determination of ovulation status for each participant was based on the detection of both (1) an LH peak, (2) and PdG rise, and (3) whether the LH peak preceded the PdG rise by no more than seven days. If all criteria were met, the cycle was considered ovulatory. If not all of the criteria were met, the cycle was deemed non-ovulatory/abnormal.

The classification was carried out with R Studio (R Core Team, 2021). All scripts for data preparation, determination of ovulation status and further analyses are openly accessible at https://osf.io/cmg3r/?view_only=1f824d13a9134483940a480f5f2bf7e6.

2.5.2. Exploratory analysis: DRSP symptoms by ovulation status

As an exemplary analysis to demonstrate how ovulation status might be used in studying symptoms, we compared individual DRSP items and the DRSP sum score between participants with and without ovulatory cycles in this sample of peripubertal females. As a fluctuation-sensitive measure, the area under the curve (with respect to ground (AUCg)) was calculated for DRSP symptom ratings across the menstrual cycle (Hambridge et al., 2013; Pruessner et al., 2003). The AUCg was compared between ovulatory and non-ovulatory participants using analyses of variance (ANOVA), and analyses were further subjected to the Benjamini Hochberg procedure (Benjamini and Hochberg, 1995) to adjust for multiple comparisons. To account for cycle length variability, and therefore number of assessments, AUCg measurements were divided by the number of assessments to compare between participants. This analysis was conducted with R version 4.2.0 (R Core Team, 2021) and the script is accessible at https://osf.io/cmg3r/?view_only=1f824d13a9134483940a480f5f2bf7e6.

2.6. Part 2: Cycle phase determination

Based on the results of Part 1 and the recommendations proposed by Schmalenberger et al. (2021) for cycling adults, we propose the following recommendations for determining precise cycle phases for individuals with frequent anovulatory cycles and short luteal phases, characteristic of the first-gynecologic year. These determination methods aim to catch specific hormonal events (levels and/or acute changes) associated with the cycle phase.

Using the dates of menses onset (before and after the assessment period) and the date of the LH peak as anchor points, as proposed by Schmalenberger et al. (2021), the mid-follicular phase (very low progesterone, rising estrogen) consists of days −7 to −3 before the LH peak, and the periovulatory phase consists of day −2 to +1 surrounding the LH peak (strong rise and fall of estrogen, slight increase of progesterone). The perimenstrual phase (falling and low estrogen and progesterone) can be classified as the four days surrounding the onset of menses (day −2 to +2), including one fewer day premenstrual than the recommendations for adults, due to shorter luteal phases. To account for typically shorter luteal phases in adolescents (median in this sample: 8 days, compared to 13.3 days currently proposed for adults; see Fehring et al., 2006, for meta-review), we recommend defining the length of the mid-luteal phase as 35% of the individual luteal phase length (rounded to full days). This individualized approach avoids relatively long luteal phase lengths that might wrongly include the periovulatory or the premenstrual phase. We recommend determining the halfway point between the LH peak and the onset of menses and anchoring the predefined phase length around this mid-point. If no halfway point can be determined since the luteal phase has an even number of days, we recommend using the days closer to menstruation to avoid including periovulatory hormone fluctuations. If an assessment day can be classified as falling into both the mid-luteal and perimenstrual (or periovulatory) phase, we recommend assigning the day to the perimenstrual (or periovulatory) phase. A mixed model approach with phase as a repeated variable (mid-follicular, periovulatory, midluteal, perimenstrual) was used to examine phase differences in mean hormone levels between ovulatory and non-ovulatory/abnormal cycles. Thus, the measure for each cycle phase was calculated with the mean of all daily values in that phase.

The detailed procedure can be derived and replicated from the openly accessible R-script (https://osf.io/cmg3r/?view_only=1f824d13a9134483940a480f5f2bf7e6).

3. RESULTS

3.1. Sample characteristics

Following screening and enrollment, 49 participants were enrolled in the study. Of those, 11 were excluded from the menstrual cycle analyses because they provided fewer than 75% of daily urine samples between day six and the end of the investigated cycle (7/11 participated in the shorter, unmodified protocol and 4/11 participated in the longer, modified protocol). Three participants dropped out of the study, either before starting the daily measures (n = 1), soon after starting the daily measures (n = 1) or in the middle of participation (n = 1). Further, 4 participants were excluded because the self-reported and parent-reported menarche dates differed by one or more years or the participant was more than 15 months post-menarche. One participant that fulfilled inclusion criteria at screening could not be scheduled for the enrollment session within 15 months after menarche. As such, 32 participants provided sufficient cycle data and were included in the present analysis.

The mean age of participants was 13.3 years old (SD = 0.78), the mean gynecological age (time since menarche at enrollment) was 9.66 months (SD = 4.20) (Table 2). Demographic characteristics did not differ significantly between participants with ovulatory and non-ovulatory/abnormal cycles.

Table 2.

Demographic Characteristics

	Total (N=32)	Ovulatory (N=13)	Non- ovulatory (N=19)
	Mean ± SD [Min, Max] or n (%)
Age	13.3 ± 0.78 [11.9, 14.8]	13.3 ± 0.81 [12.3, 14.7]	13.3 ± 0.81 [11.9, 14.8]
Gynecological age (in months)	9.66 ± 4.20 [0.77, 15.7]	10.4 ± 3.51 [3.13, 15.2]	9.10 ± 4.65 [0.77, 15.7]
Age at menarche	12.0 ± 0.84 [11.0, 14.0]	11.8 ± 0.99 [11.0, 14.0]	12.1 ± 0.73 [11.0, 13.0]
BMI	19.9 ± 3.57 [14.6, 30.2]	20.4 ± 4.04 [14.6, 30.2]	19.5 ± 3.26 [15.0, 27.3]
Member of LGTBQ community
No	27 (84.4%)	12 (92.3%)	15 (78.9%)
Yes	3 (9.4%)	0 (0%)	3 (15.8%)
Race
Black or African-American	5 (15.6%)	3 (23.1%)	2 (10.5%)
Asian	1 (3.1%)	1 (7.7%)	0 (0%)
White/Caucasian	21 (65.6%)	7 (53.8%)	14 (73.7%)
More than 1 race or other	5 (15.6%)	2 (15.4%)	3 (15.8%)
Ethnicity
Hispanic and/or Latina	2 (6.3%)	1 (7.7%)	1 (5.3%)
Not Hispanic or Latina	29 (90.6%)	12 (92.3%)	17 (89.5%)
Highest education of parent
Associate’s degree	3 (9.4%)	0 (0%)	3 (15.8%)
Bachelor’s degree	10 (31.3%)	5 (38.5%)	5 (26.3%)
Completed some postgraduate	4 (12.5%)	1 (7.7%)	3 (15.8%)
Law, or medical degree	2 (6.3%)	1 (7.7%)	1 (5.3%)
Master’s degree	5 (15.6%)	3 (23.1%)	2 (10.5%)
Other advanced degree beyond a Master’s degree	7 (21.9%)	3 (23.1%)	4 (21.1%)
Estimated annual income of parent
Less than $20,000	3 (9.4%)	3 (15.8%)	0 (0%)
$20,000 to $49,999	2 (6.3%)	2 (10.5%)	0 (0%)
$50,000 to $99,999	15 (46.9%)	8 (42.1%)	7 (53.8%)
Over $100,000	11 (34.4%)	5 (26.3%)	6 (46.2%)

Note.

¹Based on menarche date reported by parent;

²reported by parent at enrollment. Income and education of other parent were missing for the majority of participants and therefore not included in this table.

3.2. Part 1: Determining ovulation status

3.2.1. Ovulation Status

LH peak detection:

After applying the seven methods to our dataset, four methods were immediately excluded because they detected LH peaks in 90 – 100% of participants, which is unlikely one year post-menarche (Apter et al., 1978; Hoff et al., 1983; Testart et al., 1981; Zhang et al., 2008). These methods also detected multiple LH peaks per cycle (on average three peaks for Hoff et al. and Testart et al.). The remaining three methods (Brown, 1977; Johansson et al., 1971; Park et al., 2007) were compared using the rating system described in the previous section (Figure 1). The interrater reliability for the doubled rating yielded moderate to almost perfect unweighted Cohen’s Kappa values (k_Brown = 0.94, k_Johansson = 0.89, k_Brown = 0.73), based on the interpretation of kappa from McHugh (McHugh, 2012). The method proposed by Park et al. 2007 yielded the highest score, detecting an LH peak if the LH concentration exceeds a 2.5-fold increase of the rolling average of the last five days. The Park et al. 2007 method detected an LH peak in 84.38% of participants.

PdG rise:

After applying three relational methods for detecting PdG rises to our dataset (Kassam et al., 1996; Sun et al., 2019b; Zhang et al., 2008), we confirmed that the majority of detected PdG rises were in the luteal phase and that no more than 80% of cycles showed a PdG rise, which would be implausible in this age group. The methods were compared using the described rating system. The interrater reliability for the doubled rating yielded moderate unweighted Cohen’s Kappa values for the methods proposed by Kassam et al. (1996) and Sun et al. (2019b) (k_kassam = 0.62, k_Sun = 0.78) and weak interrater reliability for the method proposed by Zhang etal. (2008) (k_kassam = 0.5). The method proposed by Sun et al. (2019b) yielded the most points. The method proposed by Sun et al., (2019b) detects a PdG rise if the PdG concentration exceeds a threefold increase of the mean PdG concentration in the follicular phase and accurately identified a PdG rise in 50% of participants.

Ovulation status:

Out of the participants that showed both an LH peak and a PdG rise, three did not meet the time criterion to be considered ovulatory, with two having a PdG rise before the LH peak and one having a PdG rise 30 days after the LH peak. Five participants had neither an LH peak nor a PdG rise. Except for the three participants that missed the time criterion, all other participants with no ovulation detected had only an LH peak but no PdG rise (n = 11). This resulted in n = 13 (40.62%) participants with an ovulatory cycle and n = 19 (59.38%) with no ovulation detected.

3.2.2. Menstrual cycle characteristics

Menstrual cycle characteristics of included participants are reported in Table 3. Cycle length varied between 20 and 43 days. Of note, cycles were longer (mean = 45.27 days, SD = 11.10; min: 21 days; max: 93 days; n = 11) for the participants excluded due to insufficient data.

Table 3.

Cycle Characteristics and DRSP Symptoms by Ovulation Status

	Ovulatory (N=13)	Non-ovulatory (N=19)	p-value
Biological and cycle characteristics
Cycle Length
Mean ± SD	27.8 (5.44)	27.8 (6.85)	0.979
Median [Min, Max]	27.0 [22.0, 41.0]	27.0 [20.0, 43.0]
Follicular phase length
Mean ± SD)	19.3 (4.21)
Median [Min, Max]	18.0 [15.0, 29.0]
Luteal phase length
Mean ± SD	8.54 (2.22)
Median [Min, Max]	8.00 [5.00, 12.0]
LH AUCg (Mean ± SD)	16.1 (5.50)	24.5 (37.7)	0.351
E1G AUCg (Mean ± SD)	31.1 (8.99)	30.5 (13.7)	0.896
PdG AUCg (Mean ± SD)	819 (342)	499 (297)	0.0115*

N = number, SD = standard deviation, LH = luteinizing hormone, PdG = pregnanediol glucuronide, E1G = estrone-3-glucuronide; AUCg = Area under the curve with respect to ground, measure for overall output across the cycle.

For the 13 participants classified as ovulatory, the LH peak was determined on average 9.53 days before the onset of next menses (arithmetic mean of backward count −9.54; sd = 2.22 days; min: −13; max: −6 days). Applying the classification of Sun et al. (Sun et al., 2019b), 9 participants had short luteal phases (5 – 9 days starting from the day after LH peak until the onset of menses). The other four participants had a luteal phase length between 10 and 14 days.

Visual inspection of hormone profiles revealed notable irregularities, including three participants with an elevation in PdG early in the cycle (within three days after menses onset), which was the sole peak or accompanied by a later PdG rise. Moreover, one participant showed an early (seven days after menses onset), and two participants exhibited a late LH peak (within two days before next menses onset). Two additional participants showed unusually early elevations of LH and PdG, however; upon closer inspection, the creatinine on the respective day was marked as low, therefore the creatinine-corrected concentration was overestimated, and the non-corrected hormone concentrations did not differ from those of the surrounding days.

3.2.3. Exploratory analysis: DRSP symptoms by ovulation status

Non-ovulatory participants exhibited greater difficulty concentrating (AUCg) (F(1,30) = 4.37, p = .045, η_p² = .13) compared with ovulatory participants; however, this result did not survive the Benjamini Hochberg adjustment for multiple comparisons (p = .29). No additional DRSP symptoms differed by ovulation status, (refer to Appendix A for DRSP symptom results).

3.3. Part 2: Cycle phase determination

Cycle phases were determined according to the previously described procedure for all ovulatory participants. For comparison purposes, cycle phases were additionally determined for participants with non-ovulatory/abnormal cycles using counting methods (Figure 3). E1g (F(3.28)=15.05, p<.0001), PdG (F(3,20.5)=8.91, p=.0006) and LH (F(3,29.5)=9.19, p=.0002) levels differed by phase, with PdG levels greater overall in ovulatory participants (Group: F(1,29.8)=14.27, p=.0007). A Group X Phase interaction was found for E1g (F(3,28)=5.44, p=.005). such that E1g was greater for ovulatory participants during mid-follicular (F(1,25)=6.08, p=.021, η_p²=.20) and periovulatory (F(1,30)=12.89, p=.001, η_p²=.30) phases. Additionally, PdG (Group X Phase: F(3,20.5)=7.36, p=.002) and LH (Group X Phase: F(3,29.5)=17.01, p<.0001) profiles differed by ovulatory group status (Figure 2). Specifically, planned pairwise comparisons demonstrated that ovulatory participants exhibited greater PdG during mid-follicular (F(1,25)=36.50, p<.0001, η_p²=.59) and mid-luteal phases (F(1,30)=21.10, p<.0001, η_p²=.41), and had greater LH levels during periovulatory (F(1,30)=13.45, p=.001, η_p² = .31) and mid-luteal (F(1,30)=5.60, p=.025, η_p²=.157) phases compared with anovulatory participants.

Figure 3. Overview of Ovulation Detection and Phasing Procedure.

LH = luteinizing hormone.

^ain non-ovulatory cycles, no ovulation occurs. Therefore, the typical hormonal fluctuations across the cycle do not occur and determining “phases” is only useful when phases of ovulatory cycles need to be contrasted with days in anovulatory cycles, that are similar in length and time-point of assessment (as done in this study)

Figure 2. All Participants’ Hormone Concentrations By Cycle Phase and Ovulation Detection.

LH = luteinizing hormone, PdG = pregnanediol glucuronide, E1G = estrone-3-glucuronide. Boxplot shows the median, the 1^st, and the 3^rd quartile. The whisker extends from the hinge to the largest value no further than 1.5 * inter-quartile range from the hinge. For participants that were ovulatory, cycle phases were determined in relation to the day of ovulation. For non-ovulatory participants, cycle phases were determined according to counting methods as proposed by Schmalenberger et al. (2021).

4. DISCUSSION

4.1. Summary of results

Using daily urinary metabolites of ovarian hormones and prospective ratings of menstrual-related psychological symptoms, the present study identified the most effective methods for classifying peripubertal ovulation status based on the occurrence of a LH peak and PdG rise. After comparing ten relational methods (based on individual mean hormone concentrations and not absolute thresholds), those proposed by Park et al. (2007) for an LH peak and Sun et al. (2019b) for a PdG rise demonstrated the best fit. These methods were applied to determine menstrual cycle phases and to compare DRSP symptom patterns tied to hormone changes between ovulatory and non-ovulatory participants.

Using the methods proposed by Park et al. (2007) and Sun et al. (2019), 40.6 % of participants had an ovulatory cycle, which is consistent with prior reports of the proportion of ovulatory cycles within the first gynecological year (Gunn et al., 2018). By examining individual hormone trajectories, we found short luteal phases consistent with Sun et al. (2019b) and Carlson and Shaw (2019). Ovulatory participants exhibited greater concentrations of LH and E1G during the periovulatory phase and higher PdG during the mid-luteal phase, consistent with previous results (Hambridge et al., 2013); these hormone patterns were not observed in non-ovulatory participants. Further, patterns deviating from expected hormone trajectories across the cycle were found in 23% of participants, including early PdG elevations and late LH peaks. An early PdG rise and late LH peak (immediately preceding menses onset) may have resulted from misinterpreting mid-cycle spotting as menstruation (prevalent in 4.8% of regular cycling adults (Dasharathy et al., 2012)). To account for this possibility, we assessed days of bleeding to confirm menses length of at least three consecutive days.

Non-ovulatory participants exhibited greater difficulty concentrating and a trend towards greater symptoms of depression and hopelessness, yet results did not reach significance after adjusting for multiple comparisons. With a larger sample size and stronger statistical power, future studies should continue to examine the potential influence of ovulation status on menstrual-related psychological symptoms. It is possible that heightened stress exposure influenced symptom expression in non-ovulatory participants, as stress has been shown to affect both menstrual cycle regularity (Acevedo-Rodriguez et al., 2018; Ayrout et al., 2019) and affective symptoms (Hammen, 2005). Future research is warranted to investigate the role of stress on hormone and symptom relationships and to examine biological underpinnings of potential symptom reducing effects of ovarian hormones.

4.2. Proposed guidelines and recommendations derived from the present study

Based on the results and study design challenges of the present study, the following guidelines and recommendations are proposed to facilitate further investigation of the peripubertal menstrual cycle, detection of ovulation status, and identification of more precise cycle phases. These guidelines are meant to evolve and adapt with advancements in our understanding of the peripubertal menstrual cycle and psychological characteristics. A more extensive checklist of these recommendations for study planning purposes can be found in Appendix B.

4.2.1. General recommendations for studying the menstrual cycle in peripuberty

Due to the high variability in peripubertal cycle length, we recommend specifying an assessment time frame that allows for very long and short cycles. Cycles longer than 40 days are common (28.5% in our sample before excluding participants) and almost all reviewed studies by Gunn et al. report mean cycle lengths between 32 and 45 days. Therefore, we recommend allowing at least 45 days to assess a complete menstrual cycle. It is also essential to differentiate between mid-cycle spotting and menstruation by determining the length (through asking daily “Did you experience menstrual bleeding or spotting today?” and intensity (i.e. using pictograms of pads as reviewed by Magnay et al. (2018)) of menses.

Further, due to irregularity in cycle lengths, variability in ovulation occurrence (41% ovulatory in our sample, 0-45% ovulatory as reviewed by Gunn et al. (2018)), and a high percentage of luteal phases shorter than nine days (69% in our sample, 22% reported for a sample up to 3.5 years post-menarche by Sun et al. (2019)), frequent measures of ovulation status are necessary. We recommend daily measurements to avoid missing ovulation or to be confident that ovulation did not occur. Specific methods to determine ovulation status are discussed below. Daily symptom assessments are also recommended when comparing psychological measures between precise cycle phases. If feasible, the assessment of multiple cycles is recommended to reveal stable cycle characteristics (Schmalenberger et al., 2021) and examine the effects of irregular or anovulatory cycles on subsequent cycles (Hambridge et al., 2013). This is particularly important for studies focused on cyclical affect changes, since retrospective (single timepoint) reports of cyclical affective changes have proven vulnerable to false positive reports of premenstrual affective changes (reviewed in Eisenlohr-Moul et al.(2021)).

4.2.2. Determine ovulation status and cycle phases

This study highlighted the variety of ovulation detection methods that produced profound inconsistencies in the detection of ovulatory cycles in peripuberty, consistent with previous research in adults (Carlson and Shaw, 2019; Lynch et al., 2014). As such, recommended methods for adults by Schmalenberger et al. (2021) are not applicable for this gynecological stage. While the gold standard for precisely assessing ovulation is a vaginal ultrasound, this method is not easily accessible or feasible for most longitudinal psychological research studies.

With this study we aimed to explore other available methods to determine ovulation status (measurement of ovarian hormones, LH test, counting methods) to provide recommendations for their accuracy. An overview of these recommendations can be derived from figure 3. Of note, these recommendations are meant to evolve as research progresses.

When possible, we strongly recommend measuring ovarian hormones (preferably daily) to assess ovulation status. Therefore, it is advised to use relational methods instead of absolute thresholds to determine the occurrence of LH peaks and PdG (or progesterone) rises in relation to each individual’s mean hormone concentrations. Different methods derived from the literature can be found in table land their application is presented in the open-source R script (https://osf.io/cmg3r/?view_only=1f824d13a9134483940a480f5f2bf7e6). We developed a rating system to find the best fitting method to a sample, which is described in section 2.3.1. This can be applied and adapted to other data sets and studies. In the present study, the methods by Park et al. (2007) and Sun et al. (2019b) yielded the most accurate determinations of PdG rise and LH peak (described in 2.2.3). That said, alternative methods may be superior for other age groups, assessment specimen, or hormone analysis methods. In the present study, dried urine samples provided a feasible and precise method for assessing hormone patterns and determining ovulation status, consistent with Carlson and Shaw (2019). We recommend using creatinine corrected hormone concentrations to account for hydration status, which can alter urinary biomarker concentrations. Evaluating samples with low creatinine concentrations (below 0.1mg/mL) is essential because they may skew corrected hormone levels. If frequent hormone collections are available, visually inspecting hormone curves provides a valuable method for detecting and addressing unusual patterns, which are common in peripuberty. The open-source R script provides a subscript to plot and save individual hormone curves along with detected LH peaks and PdG rises.

LH testing to determine ovulation status is advised when hormone assessment is impossible; however, this is less reliable as LH testing alone cannot indicate if a PdG rise is present. In our sample, the sole measurement of LH peaks (using the Park et al. 2007 method) would have resulted in falsely classifying n = 11 (34%) participants. Specifically, those with an LH peak but no PdG rise (assuming that all LH peaks detected through daily hormone measured would have been detected by an LH test). To assess the validity of LH testing, we compared LH tests and ovarian hormone measurements in another sample of older adolescents. Twelve participants used ovulation detection kits (Clearblue, sensitivity 40mIU/ml) for 10-18 days starting on days 3-5 (depending on age) and assessed dried urine every second day across one or two cycles (13 cycles available in total). Comparing ovulation status based on LH tests and the determination method based on Sun et al and Park et al., 54% of cycles (7/13) showed consistent results for the two methods (ovulatory vs. non-ovulatory/abnormal). One participant showed an LH peak with both methods but did not show a PdG rise and was therefore classified inconsistently. Three participants (25%) had a positive ovulation test, yet did not show a LH peak with dried urine testing. Two participants did not have a positive LH test but showed a LH peak and PdG rise in dried urine testing. In summary, urine LH tests with a threshold of 40 mIU/ml correctly classify ovulation in more than half of participants and can be considered superior to simple counting methods (as long as the LH testing covers the majority of the cycle); however, high frequency LH and PdG sampling remain superior for detecting ovulation in peripubertal females. Still, for most studies, ovulation detection with LH tests is more feasible for most research studies than daily hormone assessment, and thus, establishing the validity and the optimal sensitivity of LH tests for adolescent participants are important areas for future research.

When LH testing and frequent hormone samples are not available, counting methods can be used to estimate ovulation date and cycle phases. However, it is not advised to use counting methods to identify ovulation dates, mainly because high rates of anovulatory cycles are expected. Therefore, the periovulatory phase cannot be reliably determined when only counting methods are available. Further, luteal phases can be relatively short in peripuberty (mean = 8.54 when counting the day of ovulation as day 0) compared with adults. Therefore, counting back 14 days from the onset of menses would lead to missed detection of ovulation in a large proportion of participants. In our sample, no participant showed ovulation exactly 14 days before the onset of menses. Therefore, assessment of mid-luteal phase based on only counting methods can be imprecise and might include unrecognized periovulatory days of short luteal phases. As such, we emphasize that identifying the mid-luteal phase based only on counting methods is not recommended. If conducted, it should include days very close to menses and must be interpreted with utmost caution. The perimenstrual and mid-follicular phases can be estimated using the start date of menses (before and after assessment) as anchor points (Figure 3).

4.2.3. Study design and feasibility with an adolescent sample

Overall participants were compliant, especially when parents were involved and informed about the study tasks. We aimed to make participation exciting and manageable by providing a checklist with study tasks, binder clips with the study logo, and allowed participants to use their personal smartphone device or a study-supplied tablet to complete daily symptom ratings. We maintained active communication with participants using the survey app’s encrypted messenger and logged pictures of the sample cards’ times and dates to ensure accurate timing of hormone sampling.

It is recommended and necessary to have parents fill out the screening questionnaire, and both parents and participants complete the enrollment forms to identify discrepancies (e.g., date of menarche). In the present study, the majority (61%) of participants reported a menarche date differing from their parent’s response, some by over a year. This deviation may have been caused by a falsely reported date or miscommunication between parents and the participant. To prevent these discrepancies in future studies, it is advised to ask both the parent and the participant for important information such as menarche date, and provide an opportunity for discussion so that a consensus can be reached and the date can be corrected. Further, it is recommended to systematically assess whether participants have told their parents about the onset of menstrual cycles at the time of menarche. Lastly, we recommend assessing contextual information about the time point of menarche (e.g., age and season it occurred, grade level, on a memorable family trip etc.).

4.3. Limitations

Despite the dense collection frequency of hormones and DRSP symptoms during peripuberty in the present study, results should be interpreted in light of several limitations. The focus of this investigation was to evaluate and improve methods for studying the peripubertal menstrual cycle. The protocol was refined as the study progressed to account for challenges that emerged, including extending the collection period from 28 to 48 days to capture the majority of menstrual cycle days.

While the sample size did not allow for including predictors and risk factors or focusing on individual patterns of hormone-sensitivity in more complex analyses, it was sufficient to conduct exploratory analyses as an example of the proposed recommendations. The results of these analyses need to be further investigated with larger sample sizes in future studies. Despite cycle irregularity and long cycle lengths in peripubertal participants, it is recommended to collect data for more than one menstrual cycle when investigating cycle effects. This would be important for examining individual cycle regularity and hormone trajectories of surrounding cycles.

Lastly, we did not validate the proposed ovulation detection methods with ultrasound measurements, which has been done previously (Sun et al. 2019). Further research is needed to compare ultrasound to the different LH peak/PdG rise detection methods, as highlighted by Lynch et al. (2014).

4.4. Future research

Developing ovulation detection methods (algorithms to detect peaks and optimal LH test kits) and validating methods with ultrasounds on large samples of peripubertal participants is one important step for future research. Additionally, the results derived from this study indicate that ovulation status is relevant to assessing hormone-affect relationships, suggesting future research possibly including the impact of stress is warranted. Further, previous research suggests that anovulatory cycles might be affected by the hormone milieu of previous cycles and influence following cycles (Hambridge et al., 2013). Thus, it would be important to investigate longer periods of cycles to examine the interaction of these varying hormone patterns.

The recommendations provided herein can be used to investigate biological underpinnings of the interaction of hormones and affective symptoms across the menstrual cycle to determine which symptoms are especially susceptible to hormone change.

Lastly, these guidelines provide an orientation for how to address and investigate other phases across the reproductive lifespan and populations that show greater proportion of anovulatory cycles, including the menopause transition. Specifically, the process to identify the most precise methods from the available literature can be applied to other samples. However, the specific peak detection methods most suitable to other samples require further investigation and validation.

4.5. Conclusion

Given the challenges of assessing the adolescent menstrual cycle with often anovulatory and irregular cycles, a critical research gap on reproductive mental health in adolescents has developed. The results and the derived recommendations of this study sought to inform future research on investigating menstrual cycle characteristics in samples where anovulatory cycles with abnormal hormone patterns that do not map onto the typical cycle are expected, including peripuberty and perimenopause. Results demonstrate that it is possible to investigate the menstrual cycle despite its irregularity in peripuberty. Further, understanding the impact of the menstrual cycle and associated hormone patterns on affective symptoms will assist in identifying early risk factors for hormone related psychopathology during peripuberty, a critical developmental window for prevention and intervention efforts.

HIGHLIGHTS.

• Ovulation determination methods for adults cannot be applied to adolescents

• Relational methods to determine LH peak & PdG rise are essential for ovulation detection

• Daily hormone measurements are recommended, LH testing needs to be modified to sample

• Counting methods need to be used with caution in samples with high rates of anovulation

• Recommendations for menstrual cycle research in peripuberty are discussed

Appendix

Appendix A:

Comparison of affective symptoms by ovulatory status

	not ovulatory (N=19)	ovulatory (N=13)
Feeling depressed
AUCg (Mean ± SD)	2.09 ± 2.10	1.31 ± 0.375
Feeling hopeless
AUCg (Mean ± SD)	1.53 (0.792)	1.12 (0.308)
Feeling anxious
AUCg (Mean ± SD)	1.85 (0.919)	1.49 (0.563)
Moodswings
AUCg (Mean ± SD)	1.76 (0.822)	1.40 (0.535)
Rejection sensitivity
AUCg (Mean ± SD)	1.76 (0.972)	1.37 (0.576)
Feeling angry
AUCg (Mean ± SD)	2.05 (1.12)	1.65 (0.630)
Low interest
AUCg (Mean ± SD)	1.48 (0.548)	1.37 (0.664)
Difficulty concentrating
AUCg (Mean ± SD)	2.00 (1.01)	1.37 (0.461)
Interpersonal conflict
AUCg (Mean ± SD)	1.54 (0.736)	1.32 (0.390)
Feeling overwhelmed
AUCg (Mean ± SD)	1.72 (0.814)	1.34 (0.491)
DRSP Sum score1
AUCg (Mean ± SD)	16.0 (6.90)	12.30 (3.31)

Appendix B:

Checklist for studying the menstrual cycle in adolescents

Recommended procedures for studying the peripubertal menstrual cycle
Assessing menarche	Assess menarche dates from both participant and parent. Ask participant if they told their parents that they started their period at that time or not/later. Assess contextual information at time of menarche (age it occurred, season of the year, nearest holiday) to confirm date accuracy. If parent and participant disagree by more than 3 months, facilitate a discussion to attempt to come to a consensus.
Length of assessment period	Allow at least 45 days of daily sampling for assessment of one menstrual cycle.
Sampling frequency	Ideally, collect hormone samples daily to aid formation of habit and analyze hormone peaks. If that is not feasible, collect hormones in high frequency (e.g. every second day) or use ovulation tests (LH strips) daily starting within the first week after menses onset.
Identifying menses	Confirm multiple days of bleeding before determining menses onset and assess heaviness of bleeding (bleeding vs. spotting) to exclude possibility of misinterpreting mid-cycle spotting as menses.
Measuring ovulation status	When vaginal ultrasound is inaccessible or unfeasible, ovulation status may be determined from daily hormone measurements based on the occurrence of an LH peak and PdG rise in relation to the rest of the cycle. Relational methods and not absolute thresholds are important in peripuberty. Methods are provided in the manuscript.
Determining cycle phase	The date of the onset of menses and the date of the LH peak may be used as anchor points to determine cycle phase as such:      Perimenstrual: Days −2 to +2 before and after menses onset.      Mid-follicular: Days −7 to −3 before LH peak.      Periovulatory: Days −2 to +1 before and after LH peak.      Mid-luteal: Individual phase length (35% of luteal phase) around midway point between LH peak and onset of menses      If no ovulation measure is available:      Perimenstrual: Days −2 to +2 before and after menses onset.      Mid-follicular: +4 to +7 after menses onset.      Periovulatory: not recommended.      Mid-luteal: −6 to −3 before onset of menses.      ➔ Attention: use with caution when no ovulation measure is available      See manuscript for more details
Daily compliance checks	Participants complete a daily survey, entering the time and date of their daily sample for study staff to monitor compliance.
Sample assessment	Participants receive instructions for sample collection and storage, and a study checklist to keep track of daily samples and surveys. They receive a large binder clip to dry urine card in open air before storage in freezer. To increase compliance, participants send a picture of sample card at the time of sample assessment to the study team. Urine cards are labeled with freezer-save print or marker.

Appendix C:

Validation studies comparing liquid and dried urine methodologies.