Salivary/Serum Progesterone Ratio Differs Between Menstrual Cycle Phases but Not Between Populations: Implications for Health, Reproductive, and Behavioral Research

ABSTRACT

Objectives

Many investigations of human health, behaviors, and adaptations require an indicator of ovarian cycle functioning as a causal, outcome, or confounding variable in the study design and analyses. Because the dynamic fluctuations in cycle hormones can rarely be adequately characterized by a single measurement, but repeated blood sampling can be onerous, salivary free progesterone (P_Free‐SAL) concentration is widely used in both clinical and research contexts as an alternative to total progesterone concentration in venous blood samples (P_Total‐VEN). However, some doubts have been raised about the use of P_Free‐SAL because of suggestions that Bolivian and other populations and/or individuals might differ markedly in the ratio of P_Free‐SAL to P_Total‐VEN (the apparent uptake fraction, UF). If there are such differences, several decades of comparative population research based on P_Free‐SAL would require reconsideration, and a seemingly useful tool in both clinical and research contexts would be lost or require additional extensive pre‐use evaluations. Such impacts would fall disproportionally on clinical monitoring and research studies of menstruating persons, a segment of the population that has long been underrepresented in research and clinical trials, especially in low resource conditions. Therefore, we tested three hypotheses: (H1) UF differs by ovarian cycle phase; (H2) UF differs in Bolivian women from that of non‐Bolivian women; and (H3) within a population, UF is consistently higher or lower in some individuals than in most others.

Methods

We collected mid‐follicular and mid‐luteal near‐concurrent samples of venous blood and saliva from 36 healthy premenopausal Bolivian women. P_Total‐VEN and P_Free‐SAL were measured using commercial enzyme immunoassays. To test the study hypotheses, we used graphical and statistical methods to analyze these new data and to analyze data from several previously published studies.

Results

In our study sample of Bolivian women, P_Free‐SAL and P_Total‐VEN concentrations (n = 66 pairs) were significantly and highly correlated (Spearman's rho = 0.858; mixed model: intercept = 77.4 pmol/L [(p < 0.001), β = 0.0191 (p < 0.001)]). An individual's follicular‐phase UF and luteal‐phase UF were not significantly correlated (rho = −0.19, p = 0.462). Median UF equaled 8.1% for follicular and 2.3% for luteal phase pairs and were comparable to published values for other populations.

Conclusions

Hypothesis 1 was supported. Consistent with prior reports for other populations, in these Bolivian women UF was higher and more variable in the follicular than in the luteal phase. The source(s) of phase‐associated variation in UF deserves additional study, particularly the dynamic relationship to different conformers of corticosteroid‐binding globulin (CBG). Hypothesis 2 was not supported. Paired P_Free‐SAL and P_Total‐VEN were highly correlated, and UF in these Bolivians was comparable to published values for other populations. Hypothesis 3 was not supported. There was no evidence that some individuals have consistently higher (or lower) UF than most other persons. In sum, these findings do not support the suggestions that the physiology underlying the relationship between P_Free‐SAL and P_Total‐VEN differs substantially and inexplicitly between populations and individuals. These results also reinforce the critical roles of fastidious attention to sample collection and handling, judicious assessment of assay results, and appropriate statistical methods when using ovarian steroid data in any project. We suggest some guidelines for meeting these requirements. Used with due consideration for its advantages and limitations, P_Free‐SAL reliably tracks P_Total‐VEN during the menstrual cycle and is a useful option in the biomarker toolkit. Just as it is costly to continue our work with tools not up to the task, so is it costly to discard useful tools without good reason. The development (and improvement through replication) of a robust toolkit for assessing changes in and the impacts of menstrual cycle hormones is foundational to reducing gender‐based health disparities. (The linked file listed below under “Supporting Information” presents these findings in Spanish).

Scatterplot of paired measurements of P_Free‐SAL (y‐axis, log scale) versus P_Total‐VEN (x‐axis, log scale) for each of our 36 Bolivian study participants. Dotted diagonals represent lines of constant uptake fraction (UF = P_Free‐SAL/P_Total‐VEN) from 0.2% to 100%. The solid red line represents the linear regression model for luteal phase pairs (red dots), which tend to monotonically increase together, hence falling within a relatively narrow band of UF values at the highest serum values. The distribution of the follicular P_Free‐SAL–P_Total‐VEN pairs (blue triangles) is more globular and spans a wider range of UF values than the distribution of the luteal pairs. Similar data distributions have been observed in non‐Bolivian populations (e.g., New Zealand), refuting the hypothesis that UF in Bolivian women differs from that of non‐Bolivian women.

Abbreviations

E2

estradiol

P

progesterone

P_Free‐SAL

free (not bound to carrier proteins) progesterone in saliva or a saliva sample

P_Free‐VEN

free progesterone in circulation or in a venous blood sample (plasma or serum)

P_Total‐DBS

total progesterone concentration in a dried blood spot sample (from a fingertip prick)

P_Total‐VEN

total (free plus bound) progesterone in circulation or in a venous blood sample

UF

apparent uptake fraction = P_Free‐SAL/P_Total‐VEN

UF_FOL

UF during the follicular phase

UF_LUT

UF during the luteal phase

1. Introduction

Many investigations of human health, behaviors, and adaptations require an indicator of ovarian cycle functioning as a causal, outcome, or confounding variable in the study design and analyses. The belated recognition that females ¹ were severely underrepresented in biomedical studies prompted passage of the U.S. National Institutes of Health (NIH) Revitalization Act in 1993, which mandated the inclusion of female participants in most NIH‐funded research on humans (Mastroianni et al. 1994; NIH 2024). For similar policies elsewhere see, for example, Government of Canada IAP on RE (2023) and Klinge (2008).

Although there has been progress, 30 years later biases in sex and gender inclusion persist worldwide at every level of research from the bench to the community (Ah‐King 2022; Arnegard et al. 2020; Barr et al. 2024; Beery and Zucker 2011; DuBois and Shattuck‐Heidorn 2021; Geller et al. 2018; Mazure and Jones 2015; Orr et al. 2020; Wilson et al. 2020). This imbalance is exacerbated for those persons who experience health disparities (i.e., are at greater risk of poorer health and higher mortality compared to the general population) because of their ethnicity/race, low income, social status, rural residency, or other factors (Thomson et al. 2006).

In many (arguably most) studies that have included females, either the possible effects of changes in cycle hormones on outcome variables are ignored entirely or data collection is restricted to the early follicular phase (when the principal ovarian steroids, progesterone [P] and estradiol [E2], are low) (Shea and Vitzthum 2020; Wenner and Stachenfeld 2020). Many studies collect only a single venous blood sample to ascertain whether a cycle hormone concentration falls within some expected range. Such research strategies are insufficient for assessing the roles of ovarian hormones on health and well‐being during at least 75% (3 of every 4 weeks) of the roughly 40 years of menstrual cycling experienced by half of the human population.

Addressing these inequities is no simple task. The menstrual cycle is not a clock (Vitzthum 2009; Vitzthum et al. 2021). The natural variability in follicular and luteal phase durations between cycles and between persons (Figure 1) makes it difficult to accurately predict the timings of ovulation, peak preovulatory E2, or the peaks of mid‐luteal E2 and P. Hence, a single sample collected on a day presumed to represent one of these fleeting states will not necessarily be accurately timed for capturing the true hormonal concentrations at these commonly defined transitions in the cycle. Because the body dynamically responds to internal and external cues (Vitzthum 2001, 2008, 2024), reliance on a single blood sample, even judiciously and fortuitously timed, is typically not up to the task of adequately characterizing ovarian functioning (Fujimoto et al. 1990).

FIGURE 1.

Natural (“normal”) variation in serum progesterone concentration in adult premenopausal women. Inter‐cycle variability (i.e., within‐woman or intra‐woman variability) for ovulation (dark orange) and next menses (dark blue) are the 95% prediction intervals for the timing of these events in any single woman, assuming an inter‐cycle average duration that is equal to the population average. Inter‐woman (between women) variability for ovulation (orange) and next menses (blue) are the 95% prediction intervals for the timing of these events in the overall population. Hormone levels represent usual ones [and are], not necessarily related to what is healthy. Hormone ranges vary between persons at the same biological stage of the menstrual cycle. The actual timing (days from menstruation) of that biological stage varies between cycles and between women. The ranges denoted by biological stage are the 90% prediction intervals for hormone levels for women at the same biological stage (dark green). Inter‐woman variability (between women, yellow) are the up to 95% prediction intervals for hormone levels in the overall population. Source: Adapted from: https://commons.wikimEdia.org/wiki/File:Hormones_estradiol,_progesterone,_LH_and_FSH_during_menstrual_cycle.png.

Ovarian hormone concentrations are best captured by serial sampling. Because serial venous blood collection is logistically demanding and understandably onerous for most persons, methods to measure hormones in urine, saliva, and fingertip blood drops (dried blood spots, DBS) have been developed. The choice of a biomarker depends on several considerations, including the research question, available facilities, and the study population. There are various biomarkers, each with advantages and limitations, that meet these differing needs (see, e.g., Leonard (2021), Snodgrass (2022), and the papers in each issue; Gröschl (2017), McDade (2014), Vitzthum (2021), and Worthman and Costello (2009)).

The concentration of free progesterone in saliva (P_Free‐SAL) is one widely used biomarker (Figure 2). P_Free (P not bound to the carrier proteins, corticosteroid‐binding globulin [CBG], and albumin) is the biologically active (i.e., is available to the tissues) portion of total P (P_Total). Because P_Free diffuses rapidly from blood into saliva (Bolaji 1994; Vining and McGinley 1986), P_Free‐SAL is a credible proxy for (but does not equal) P_Free in venous blood (P_Free‐VEN) (Blom et al. 1993; Delfs et al. 1994; Evans 1986; Laine and Ojanotko 1999; Wang and Knyba 1985). However, P_Free‐VEN is not routinely measured because the protocols require considerably more labor and expense than does measuring P_Total in a sample of venous blood (P_Total‐VEN), and because P_Free‐VEN is not commonly used for clinical purposes (Read 1989). Therefore, the validity of P_Free‐SAL as a useful biomarker has been evaluated, with few exceptions, by comparing the concentrations of P_Free‐SAL and P_Total‐VEN. ^{2 , 3}

FIGURE 2.

Progesterone pathways from circulating venous blood to saliva. In circulation, most P molecules are bound to a carrier protein (CBG or albumin); only unbound P_Free‐Ven diffuses passively, requiring about 1 min for unidirectional passage from blood into saliva. Saliva may be contaminated by blood (from gums or other mouth injury/disease) and/or mouth debris (e.g., food, tobacco). If present, such contamination typically raises the measured P_Free‐SAL by several times the concentration of P_Free‐SAL that enters saliva via passive diffusion.

Some investigators have questioned the use of P_Free‐SAL to compare ovarian functioning across populations and individuals because of hypothesized, but unspecified, atypical secretory mechanics in Bolivians (Chatterton et al. 2006) and perhaps also in some U.S. women who had self‐identified as either Japanese or Caucasian ethnicity (Konishi et al. 2012). The authors of these two independent studies suggested that the ratio P_Free‐SAL/P_Total‐VEN (for convenience, referred to here as the apparent uptake fraction [UF]) is too variable, for whatever biological or methodological reasons, for P_Free‐SAL to be a useful biomarker of ovarian functioning for at least some persons and/or populations. Their conclusions differed markedly from previously published validation studies of P_Free‐SAL.

Many research groups have evaluated paired P_Total‐VEN and P_Free‐SAL measurements from near‐concurrently collected blood and saliva samples and have reported a high correlation between the two biomarkers (Appendix A; discussed in Section 1.2). Of those studies that also reported UF, follicular‐phase (UF_FOLL) values are typically higher than luteal phase (UF_LUT) values; phase‐specific UF values are roughly similar across studies. However, it is much rarer to compare paired P_Total‐VEN and P_Free‐SAL measurements in those populations that are culturally averse to having any blood drawn and/or in those research settings lacking the requisite facilities and personnel for venipuncture collection and subsequent storage of blood samples. Under such circumstances, and in light of the high correlations reported in laboratory validation studies, forgoing a comparison of paired P_Total‐VEN and P_Free‐SAL in novel contexts has not previously prompted much concern about the validity of P_Free‐SAL as a proxy for P_Total‐VEN.

On the one hand, it is a reasonable assumption until shown otherwise that the biochemistry of P_Free diffusion from blood to saliva (Figure 2) is comparable across human populations. As such, it appears justified to use protocols that have been laboratory validated in one population for studies in other populations.

On the other hand, if (as was argued by Chatterton et al. (2006) and Konishi et al. (2012)) the diffusion of P_Free (and perhaps other analytes) consistently and significantly differs between some populations and/or some persons, several decades of community‐based comparative research relying on P_Free‐SAL measurements would require reassessment.

Furthermore, the future use of serial saliva sampling to evaluate ovarian functioning in study participants would, at the least, require additional pre‐study validation that the UF is sufficiently similar in prospective participants and/or populations for reliably evaluating the study questions, or that UF variation can be suitably accounted for in analyses. If not, there would be fewer tools for serial measurements throughout the cycle of hormonal changes and their consequent effects on health and behaviors.

1.1. Study Hypotheses

To evaluate the utility of P_Free‐SAL as a biomarker of human ovarian functioning, we present new data from 66 sets of near‐concurrent saliva and venous blood samples collected in a research laboratory from 36 Bolivian women. We tested three hypotheses regarding the UF for progesterone (i.e., the ratio of P_Free‐SAL/P_Total‐VEN).

1. UF differs by ovarian cycle phase (i.e., follicular vs. luteal phases).

2. UF differs in Bolivian women from that of non‐Bolivian women.

3. Within a population, UF is consistently higher (or lower) in some individuals than in most others.

1.2. Prior Evaluations of P_{Free ‐SAL} as a Biomarker for Circulating Progesterone

A search for published and gray literature that reported assessments of the relationship of P_Free‐SAL to P_Total‐VEN yielded the list of studies in Appendix A. Notably, only a few of these are for populations outside the United States and Europe. However, not all validation studies are necessarily published, perhaps particularly those that are done in a different population but are consistent with previously published validations of the same (or near same) protocols.

A principal goal motivating the development of salivary steroid assays in the late 1970s was the need for minimally invasive alternatives to blood collection, thus facilitating serial monitoring of patients and the effects of treatments. Moreover, in many populations there were (and still are) strong aversions to having blood drawn, and/or a lack of facilities and trained personnel, which limit both health care and health research. International organizations (e.g., World Health Organization [WHO]) as well as research units at health centers and universities were involved in the development of various alternatives to collecting venous blood for measuring analytes of interest (International Atomic Energy Agency (IAEA) 1982; Read et al. 1984; Sufi et al. 1982, 1985).

The Tenovus Institute for Cancer Research in Cardiff, Wales, was among those first teams to develop P_Free‐SAL collection and assay protocols (R. F. Walker et al. 1978, 1979; S. Walker et al. 1981). For validation, a total of nine healthy menstruating volunteers collected daily saliva samples throughout a complete menstrual cycle (Appendix A: Studies 1 and 2). Matched venous blood samples were collected from each of these volunteers on at least 8 and up to 25 days. The correlation (r) of each volunteer's paired P_Free‐SAL and P_Total‐VEN concentrations ranged from 0.81 to 0.97 and was 0.86 for the entire sample of 143 matched pairs.

By 1980, the Tenovus Institute protocols were in use in a Bangladeshi study of reproductive functioning in impoverished women (Seaton and Riad‐Fahmy 1980). Saliva samples were collected daily for 3–4 months, a protocol not likely to have been sustained if blood samples had been collected instead. Low P_Free‐SAL throughout a cycle, indicative of anovulation or postovulatory luteal insufficiency, was common in the sample of Bangladeshi women. Such patterns can be readily missed if only one, or even a few, blood or salivary samples are collected during each cycle. An assessment of UF in Bangladeshis at that time, if done, was not reported in the available literature. Given that the investigators interpreted low P_Free‐SAL as they would have if observed in United Kingdom (UK) women, it appears the investigators assumed that the correlation between measurements of salivary and plasma P, and the UF, was similar in United Kingdom and Bangladeshi women.

Some three decades later the relationship of P_Free‐SAL to P_Total‐VEN was assessed in a new study of Bangladeshi women. Houghton (2008) compared the correlation of P_Free‐SAL and P_Total‐VEN in three samples of adult women: nonmigrant (“sedentee”) Bangladeshi, Bangladeshi who had migrated in adulthood to the United Kingdom, and Britons of European descent (Appendix A: Study 26). A single pair of saliva and blood samples was collected during the presumptive luteal phase. Participant reporting of the subsequent first day of menses revealed that some sample collections were not, in fact, during the luteal phase (predicting the timing of the luteal phase is a common difficulty in ovarian cycle research). Houghton (2008) reported several unforeseeable challenges that specifically impacted data collection from the Bangladeshi migrants, which may explain the very low correlation (r = 0.19) of P_Free‐SAL and P_Total‐VEN for this sample of women. In contrast, the correlations of P_Free‐SAL and P_Total‐VEN in the Bangladeshi sedentee and British samples were high and comparable to those observed in earlier studies (Appendix A). Moreover, UF_LUT in the British and Bangladeshi sedentee samples were similar (0.30% ± 0.18% and 0.25% ± 0.23%, respectively). These findings are consistent with the expectation that the dynamic relationship between P in blood and in saliva does not differ across human populations (or, at the least, not between Bangladeshi and British women).

For various reasons, measurements of an analyte are not necessarily comparable across laboratories (Dabbs Jr. et al. 1995). Evaluating hormone concentrations over time and across populations typically requires the use of equivalent materials and protocols. Rigorous quality controls within laboratories help to maintain consistency in measurement. With these and other requirements in mind, in the 1970s the World Health Organization (Special Program of Research in Human Reproduction) established the Matched Reagent Program in dozens of countries worldwide (Sufi et al. 1982).

As part of this program, protocols for P_Free‐SAL and salivary estradiol (E2_Free‐SAL) were evaluated in daily samples throughout a cycle from six women in each of five countries (Chile, China, India, Singapore, Thailand; Sufi et al. 1985). In this initial study (published results were not identified by country), P_Free‐SAL in three centers were comparable to those in previously published studies (the authors cited ⁴ Chearskul et al. 1982; Donaldson et al. 1984; R. F. Walker et al. 1984). In the one study center where P_Free‐SAL was unusually low, four of the six participants had been from an infertility clinic. For uncertain reasons, one center had exceptionally high values for P_Free‐SAL and E2_Free‐SAL, and variability for each steroid was several times greater than in the other centers. Nonetheless, in this center the relative changes in both steroids across cycle phases mirrored those of the other centers. The program concluded that P_Free‐SAL is a “clearly useful analytical tool, particularly for studies involving examination of endocrine status over a period of time [and] in population groups from whom regular blood samples would be difficult or impossible to obtain…” (Sufi et al. 1985).

At Harvard University (Boston, USA), Ellison adapted the Tenovus salivary progesterone protocols (R. F. Walker et al. 1979; S. Walker et al. 1981) for use in anthropological research (Ellison 1988; Ellison et al. 1986). The reported correlation of an individual's paired serum and saliva samples during “the course of the menstrual cycle” ranged from 0.80 to 0.97 in three Boston women (Appendix A: Study 20 [Lipson and Ellison 1992]). These collection and assay protocols were used in several populations (Figure 3) including Bostonians and Congolese (Ituri Forest, DRC) horticulturalists/hunter‐gatherers (Ellison et al. 1986, 1989), Nepalese farmers (Panter‐Brick et al. 1993), Polish farmers (Jasienska and Ellison 1998), and Bolivian Quechua agropastoralists (Vitzthum et al. 2000).

FIGURE 3.

Mid‐luteal progesterone (P_Free‐SAL concentration) in five population samples assayed following the same protocol in the same laboratory (Ellison et al. 1993). The left panel compares all participants in each study sample. The right panel compares similar subsamples (ages 25–35 years; cycles during the population's least stressful season) from each study (data from Vitzthum et al. (2000): table 2).

Only a few studies (Appendix A) have published (in English) evaluations of the association of P_Free‐SAL and P_Total‐VEN in populations that are outside the United States, Europe, or New Zealand: China (Liu et al. 2018; Wong et al. 1990), Thailand (Chearskul and Visutakul 1994; Vienravi et al. 1994), Iraq (Abood 2008), and Bangladesh (Houghton 2008). In every case, the correlation of P_Free‐SAL and P_Total‐VEN was greater than 0.7, except for older Iraqi women (Study 25). However, correlations say nothing about the UF (i.e., even if populations differed in UF, the correlation of population‐specific P_Free‐SAL and P_Total‐VEN would not necessarily differ across populations).

1.3. Variation of UF by Cycle Phase

Perhaps the earliest indication that UF varies by cycle phase comes from Chearskul et al. (1982) (Appendix A: Study 3). From their published summary data (their figure 8), we estimated UF_FOLL to be about 3% and UF_LUT to be about 0.9%. Subsequent independent studies likewise reported UF_FOLL to be from two to six times higher than UF_LUT (Appendix A: Studies 2, 5, 7–9, 12–14 [Bourque et al. 1986; Cedard et al. 1984; Choe et al. 1983; De Boever et al. 1986; Evans 1986; Riad‐Fahmy et al. 1987; Tallon et al. 1984; S. Walker et al. 1981; Zorn et al. 1984]). Some, perhaps much, of the variability across studies in UF likely reflects differences in calculating UF (discussed in Section 2.4 and Appendix B), selected cycle days, samples per person, and collection and assay protocols. All of these early studies were in countries where the population was predominantly of European descent (e.g., United Kingdom, United States), so it is unlikely that the study differences in UF are attributable to inherent population differences in steroid biochemistry. ⁵ Notably, a study (Appendix A: Study 23) in Thailand (Chearskul and Visutakul 1994) reported UF comparable to those reported by the United Kingdom and United States studies.

FIGURE 8.

Scatterplots of paired measurements of P_Free‐SAL (y‐axis, log scale) versus P_Total‐VEN (x‐axis, log scale) for each participant in four studies. Linear regression models (panels A and D) and linear mixed model (panel B) (see endnote 7) are shown for luteal phase pairs (excluding outliers) in each sample. x and y scales are identical across studies. Shaded areas represent values below the limit of assay sensitivity (shown if reported by study authors). Diagonal dotted lines represent lines of constant uptake fraction (UF = P_Free‐SAL/P_Total‐VEN) from 0.2% to 100%. Panel A (this study): 66 pairs from 36 Bolivians. Panel B (data from Evans 1986): 38 pairs from 4 New Zealanders. Panel C (data from Wang and Knyba 1985): 36 pairs collected on a random cycle day from 36 Britons. Study authors screened values below the sensitivity of either assay (see their figure 4 for unscreened full data set). Panel D (data from Chatterton et al. 2006): 25 pairs from 25 Bolivians, and 15 pairs from 15 Chicagoans. In panels A and B, paired samples were collected during the follicular phase (blue triangle) and luteal phase (red dot). In panel B, purple “+” indicates collection on peri‐ovulatory days. In panels A and C, green “+” indicates uncertain phase. In panel D, all samples were collected at the presumptive mid‐luteal phase (NW‐Bolivia: red dots; NW‐Chicago: black “x”). Circled symbols are P_Free‐SAL–P_Total‐VEN pairs with implausibly high UF, suggesting sample contamination.

Although these studies, individually and collectively, refuted the previously common assumption that UF is, on average, constant across the cycle, the various authors concluded that the high correlations between P_Free‐SAL and P_Total‐VEN supported the use of P_Free‐SAL for monitoring ovarian function. These findings also made evident the need to account for cycle phase when calculating the correlation of matched saliva–serum pairs. Nonetheless, depending on the research question, phase differences in UF do not lessen the utility of P_Free‐SAL as a biomarker for ovarian functioning. Changes in P_Free‐SAL follow a similar trajectory to those in P_Total‐SER. Therefore, the timing of ovulation (if it occurs) and cycle phase can be determined fairly accurately from the pattern of P_Free‐SAL variation in serial saliva samples spanning the cycle (Vitzthum 2021).

Most investigators listed in Appendix A explicitly noted the limitations of relying on a single saliva or blood sample. Although studies have concluded that neither salivary metabolism nor saliva flow rate is a source of error in P_Free‐SAL (Blom et al. 1993; Laine and Ojanotko 1999), the use of any single P measurement in diagnoses or research is problematic for several other reasons (Ansbacher 1990; Arslan et al. 2023; Fujimoto et al. 1990). Because the secretion of P is pulsatile (Delfs et al. 1994; Filicori et al. 1984; Fujimoto et al. 1990; L. L. Lewis et al. 1995; O'Rourke and Ellison 1990; Rossmanith et al. 1990; Soules et al. 1988; Veldhuis et al. 1988), a single blood draw or saliva sampling is unknowingly at the peak, nadir, or in‐between pulses. Significant diurnal variation in both P_Free‐SAL and P_Total‐VEN has also been observed in free living study participants and controlled laboratory settings (Patil et al. 2023; Rahman et al. 2019). In the body, the half‐life of progesterone is about 5 min (Kolatorova et al. 2022), and the passage from blood to saliva is about 1 min (Blom et al. 1993). Therefore, time lags between the collection of blood and saliva samples and prolonged saliva collection may yield a P_Free‐SAL measurement that is not strictly synchronous with the measured P_Total‐VEN, contributing to differences between studies in calculated UF and the correlation of P_Free‐SAL and P_Total‐VEN. Other factors that may affect measurement precision and accuracy, and hence correlation and calculated UF, include the protocols and materials for collecting and storing samples, variation in study participants' adherence to protocols, and different types of assays. Idiosyncratic and unsuitable statistical analyses also make it difficult to interpret findings and to compare outcomes across these and other studies.

These methodological differences aside, if UF does differ systematically across populations or even subpopulations, some research questions cannot be addressed with P_Free‐SAL, and much of the work and theory relying on such data would need rethinking. Furthermore, future studies might not be able to rely on this noninvasive method for adequately monitoring ovarian functioning in a broad array of investigations of the determinants of human health and adaptations. For these reasons, we conducted this study to test the three hypotheses listed above.

2. Methods

2.1. Study Sample

We recruited study participants via word of mouth and posted announcements at the Universidad Mayor de San Andrés, La Paz. All participants were native Bolivians, born and residing at ≥ 3500 m, from 20 to 40 years old inclusive, of normal weight for height, regularly cycling for at least the past 3 months, and not using hormonal contraceptives nor any other hormonal medication during the previous 4 months. Of 38 women who entered the study, 36 completed all protocols.

Each participant was interviewed in her preferred language (Aymara or Spanish) by a native speaker regarding her birthplace and date, medical and reproductive histories, residences, education, and occupations. A private training/practice session on saliva collection was done prior to the first scheduled day of sample collection. A team member demonstrated the full protocol, which was then practiced by the participant. Standard anthropometrics (weight, height, arm circumference, and four skinfolds) were measured by a single experienced physician (Dr. Hilde Spielvogel).

2.2. Sample Collection

Based on reported first day of the previous menses, two sample collections were scheduled to coincide with days approximating the mid‐follicular and mid‐luteal phases, respectively. Each collection comprised staff‐monitored participant self‐collection of saliva according to a standard protocol (Vitzthum 2021) immediately followed by venous blood draw by an experienced phlebotomist. Collection materials were made of either laboratory‐quality glass (blood) or polypropylene (saliva). On the morning of the scheduled appointment, a participant was required to refrain from eating or drinking anything other than plain water, to refrain from tooth brushing or flossing (to avoid gum irritation), and not to wear lipstick or other ointments on the lips or mouth. Upon arrival at the laboratory, she rinsed her mouth with cool clean water to remove any debris and to help constrict gum capillaries to reduce the risk of blood contamination. To avoid sample dilution, saliva collection began ≥ 20 min after rinsing. Saliva production was promoted by gently chewing a piece of inert wax, allowing saliva to drool into the collection vial. Participants avoided talking or sucking their cheeks during saliva collection, which typically required 2–5 min. A research team member monitored study participant preparation and saliva collection throughout the procedure.

Shortly after collection, blood (but not saliva) samples were centrifuged, and serum and saliva aliquots were frozen at −20°C and maintained frozen until assayed. All samples were air shipped in a single day in an insulated shipping box packed with −20°C phase‐change ice packs; a constant‐read electronic thermometer confirmed that the samples had stayed frozen throughout shipping (Thornburg and Vitzthum 2020).

For each participant, the expected date of the first day of the next menses was estimated based on prior cycle length. Participants were asked to record when the upcoming first day of bleeding actually occurred and to contact the research team. Participants were also contacted shortly after the anticipated first day and thereafter to learn the actual first day of menses. Of the 36 women, 3 were uncertain of the day of menses onset.

At their first sample collection appointments (first round, Figure 4), saliva samples from four women were contaminated by visible gum bleeding; two of these women withdrew from the study, and two elected to continue. In one other case, samples were not assayed because of a lengthy time delay between saliva and blood collection. Of 36 women who participated in the second round of collection, one saliva sample had visible blood contamination, and 2 other women had trouble producing an adequate saliva sample. A total of 66 sample pairs were assayed; all measurements were included in subsequent statistical analyses except as noted.

FIGURE 4.

Study sample composition.

2.3. Laboratory Assays

Enzyme immunoassays (EIAs) were done at the CISAB Laboratory at Indiana University, Bloomington, USA, using serum kits from American Laboratory Products Company (Salem, NH, USA) with a lower limit of detection of 0.318 nmol/L (=0.1 ng/mL) P_Total‐SER and saliva kits from Salimetrics LLC. (State College, PA, USA) with a lower limit of detection of 31.8 pmol/L (=10.0 pg/mL) P_Free‐SAL. All saliva (or serum) samples from a given participant were assayed directly on the same plate in sequential positions on the plate according to the manufacturer's instructions. For the P_Free‐SAL assay, participants were randomly allocated to one of two plates and randomly allocated to positions on the plate. For the P_Total‐SER assay, the allocation of participants and positions on the plates mirrored those of the salivary assay plates. All samples were assayed in duplicate.

A saliva–serum sample pair was designated as follicular if it was collected ≥ 15 days before, or as luteal if it was collected ≤ 14 days before, the reported first day of menses following sample collection; the luteal designation is presumptive because the occurrence and timing of ovulation in a given cycle is unknown. Of 66 serum–saliva pairs, number of follicular = 29 (median collection day = cycle day 8 [=reverse cycle day −21]), number of luteal = 34 (median collection day = cycle day 22 [=reverse cycle day −8]), and 3 were unassigned because of uncertainty in menses onset.

2.4. Analytical Approaches

P_Free‐SAL units are pmol/L (=0.3145 pg/mL); P_Total‐VEN units are nmol/L (=0.3145 ng/mL). Data were analyzed using SPSS v29.0 (IBM). Sample central tendency is reported as mean ± standard deviation (SD) or median; coefficient of variation (CV) = SD/mean. Statistical significance was set at p ≤ 0.05. Descriptive statistics were calculated for P_Free‐SAL, P_Total‐SER, UF_FOLL, UF_LUT, and participant characteristics. Possible subsample differences arising from the allocation of samples to the two assay plates were evaluated with Student's t‐test. The relationship of P_Free‐SAL to P_Total‐SER was evaluated statistically including Spearman's correlation ⁶ and hierarchical linear modeling (HLM, SPSS mixed model command). HLM accounts for the lack of independence of serial sample pairs from the same individual (repeated measurements violate standard linear regression's assumption of independence). The impact of outliers in the data set (which may arise from errors in reported days of menses, collection and/or assay protocols, sample storage, and/or sample contamination or dilution) was explicitly evaluated.

The data were also evaluated graphically. In the case of data with a large dynamic range (i.e., data for which the largest value is much larger than the smallest value, as is often the case for P_Free‐SAL and P_Total‐SER), a standard “linear” plot makes it hard to distinguish relative variation among the smaller values. A plot on logarithmic scales makes it much easier to perceive relative variation among all the values, both small and large. For both linear and log scales, scatterplots of P_Free‐SAL versus P_Total‐SER are a useful tool for understanding the joint distribution of these data. Such scatterplots can convey important information which is hidden by summary statistics or one‐dimensional plots such as histograms or cumulative distributions.

Recall that differences in laboratory protocols usually preclude direct comparison of analyte measurements across studies. However, plots using logarithmic scales allow for cross‐study comparisons of the trends and shapes in the distributions of the data. That is, suppose for example that we are comparing two studies (“A” and “B”), and that A's serum assay gives 1.25 times the result of B's serum assay for the same samples, and A's saliva assay gives 1.5 times the result of B's saliva assay for the same samples. Then in a log–log scatterplot of P_Free‐SAL versus P_Total‐SER (such as those in Figures 5 and 8), the pattern of A's and B's points will be identical, with A's pattern shifted to the right by log(1.25) and up by log(1.5) with respect to B's pattern. In a log‐scale histogram or cumulative distribution of UF (such as in Figure 9), A's pattern will again be identical to B's pattern, except shifted to the right by log(1.5/1.25).

FIGURE 5.

Uptake fraction (UF = P_Free‐SAL/P_Total‐VEN) in Bolivia sample (66 P_Free‐SAL–P_Total‐VEN pairs, 36 women). P_Free‐SAL (y‐axis, log scale) versus P_Total‐VEN (x‐axis, log scale). Diagonals represent lines of constant UF from 0.5% to 100%. For each sample pair, cycle phase is follicular (blue triangle), luteal (red dot), or uncertain (green cross). Vertical dashed line = 4 nmol/L P_Total‐VEN, above which all sample pairs have UF between 1.5% and 4.5%. Two vertical black arrows represent the phase‐specific change in UF due to the same quantity of a hypothetical contaminant in a saliva sample (discussed in text; see Table 3). Circled symbols are P_Free‐SAL–P_Total‐VEN pairs with implausibly high UF, suggesting sample contamination. Shaded areas are below the limit of assay sensitivity.

FIGURE 9.

Distribution of % uptake fraction (x‐axis, log scale) in four study samples for luteal‐phase (red) and follicular‐phase (blue) samples. Histogram (left y‐axis, bins are equally spaced in log(UF)), cumulative distribution (right y‐axis), median (◊), and IQR (|—|). Panel A (this study); Panel B (data from Evans 1986); Panels C and D (data from Chatterton et al. 2006).

2.4.1. Calculating and Comparing UF

There is no standard approach for calculating mean UF, and the most common computation (the ratio of arithmetic mean [A‐mean] of the saliva analyte to the A‐mean of the venous analyte) is not the most useful. Nonetheless, for comparison of our findings to the published literature, we calculated this ratio as has been done in most other reports. Specifically,

$$\mathrm{UF}=\frac{\mathrm{A}‐\text{mean}\left({\mathrm{P}}_{\text{Free}-\mathrm{SAL},i}\text{for}i=1,2,3,\dots ,N\right)}{\mathrm{A}‐\text{mean}\left({\mathrm{P}}_{\text{Total}-\mathrm{SER},i}\text{for}i=1,2,3,\dots ,N\right)}$$ (1)

where N equals the number of pairs of concurrently collected saliva and serum concentrations, P_Free‐SAL,i and P_{Total‐SER,i} respectively, for i = 1, 2, 3, …, N. Note that some authors have calculated this ratio using non‐paired data (i.e., all available measurements, even if one member of one or more paired samples was unavailable [e.g., lost because of contamination or assay failure]).

In Appendix B, we discuss the limitations of this common computational approach and present more suitable alternatives (e.g., geometric mean or median) for describing and comparing UF.

Regardless of which measure of central tendency is chosen, it is strongly preferable to use only paired data when computing the UF. If non‐paired data are used, then UF is inherently confounded by any inter‐cycle‐phase or inter‐individual variation in P_Free‐SAL, P_Total‐SER, and/or UF that may be present. For example, a data set with a larger fraction of luteal phase saliva samples than the fraction of luteal phase serum samples will have an upwardly biased UF. This bias is easy to envision for the extreme hypothetical case of a study sample comprising only ovulatory cycles with all serum samples collected during the follicular phase and all saliva samples collected during the luteal phase: the computed UF would be systematically biased upwards.

3. Results

Table 1 presents sample characteristics and hormone concentrations for those women who completed the study protocol (n = 36). These variables did not statistically differ between the women randomly assigned to the two plates used in each assay.

TABLE 1.

Sample characteristics and hormone concentrations for 36 participants, 66 saliva–serum pairs (29 follicular phase, 34 luteal phase, 3 phase uncertain).

		Range
Age (mean ± SD, years)	28.1 ± 6.6	20.1 to 39.8
Menarcheal age (median years)	13	9 to 19
Age at first birth (median years), n = 19	22.5	16.9 to 33.0
Number of live births (median)	1	0 to 3
Currently breastfeeding (n)	4
Education (median, years)	> 12	5 to 12+
Principal occupation
Student	53% (19)
Employed	39% (14)
Homemaker	8% (3)
Height (mean ± SD, cm)	155 ± 5.9
Weight (mean ± SD, kg)	57.5 ± 7.9
BMI (mean ± SD, kg/m2)	24.0 ± 3.6
Cycle duration (median days)	28.0	22 to 38
Follicular phase: reverse cycle day range (median)	−21	−29 to −15
Luteal phase: reverse cycle day range (median)	−8	−14 to −2
PSal‐Free (pmol/L; mean ± SD [CV])
Follicular phase	100.5 ± 50.8 [0.51]
Luteal phase	446.5 ± 277.3 [0.62]
PSer‐Total (nmol/L; mean ± SD [CV])
Follicular phase	0.989 ± 0.525 [0.53]
Luteal phase	19.14 ± 10.64 [0.56]
Correlation: follicular‐phase UF vs. luteal‐phase UF (n = 17)	rho = −0.19 (p = 0.462)

For each assay, controls (provided by the manufacturers except for pooled serum) were run in quadruplicates on two plates. For P_Free‐SAL low (137 pmol/L) and high (3121 pmol/L) controls respectively, inter‐assay CVs were 5.3% and 2.8%, and intra‐assay CVs were all < 4.1% and < 4.5%. Intra‐assay CV (duplicates of participants' samples) was 2.2%. For P_Total‐SER pooled low (6.52 nmol/L) and high (30.85 nmol/L) controls respectively, inter‐assay CVs were 13.1% and 4.4%, and intra‐assay CVs were all < 11.2% and < 4.4%. Intra‐assay CV (duplicates of participants' samples) was 4.3%.

Figure 5 is a scatterplot on log scales of P_Free‐SAL versus P_Total‐VEN for all 66 sample pairs from 36 Bolivians. Dashed diagonals represent lines of constant UF from 0.5% to 100% (mean UF reported by other researchers ranges from about 0.9% to 9% [Appendix A] depending on cycle phase and method of computation).

In our Bolivia sample, UF is between 1.5% and 4.5% for all sample pairs in which P_Total‐VEN is > 4 nmol/L (indicated in Figure 5 by vertical dashed line), and UF is < 10% for all cases of P_Total‐VEN > 1 nmol/L. A set of nine pairs (circled symbols), each with P_Total‐VEN < 1 nmol/L (for these nine, mean P_Total‐VEN = 0.536 ± 0.244 nmol/L), has implausibly high UF values (18.6%–52.5%). Eight were collected during the follicular phase and one during the very late luteal phase (red dot). Each of the nine pairs is from a different participant. The second P_Free‐SAL–P_Total‐VEN sample pair collected from each of these nine participants has biologically typical UF (1.67%–5.44%) and, collectively, much higher mean P_Total‐VEN (19.52 ± 12.16 nmol/L), typical of luteal‐phase P.

Excluding the one luteal‐phase outlier (circled red dot), P_Total‐VEN is > 1 nmol/L in all luteal phase samples, and UF_LUT is 1.5%–8.8% (median = 2.2%, G‐mean = 2.4%). Excluding the eight follicular‐phase outliers, UF_FOLL is slightly more variable (3.2%–12.6%) and greater (median = 7.3%, G‐mean = 7.0%) than UF_LUT (also see Figure 9A in Section 4.2).

The correlation of P_Free‐SAL and P_Total‐VEN is high for the total sample (Spearman rho = 0.858) and luteal‐phase pairs (rho = 0.857), but low for the follicular‐phase pairs (rho = 0.18) (Table 2). Correlation of the follicular‐phase pairs rises dramatically (rho = 0.70) once eight biologically improbable outliers are excluded from the analysis. (For comparison, Table 2 also shows results from two other studies that are discussed in Section 4.2.)

TABLE 2.

P_Free‐SAL versus P_Total‐VEN correlations for Bolivia and Chicago samples in three data sets.

	Bolivia samples	Chicago samples
This study	This study
PFree‐SAL vs. PTotal‐VEN (all pairs)	rho = 0.858 (n = 66)
PFree‐SAL vs. PTotal‐VEN (all luteal‐phase pairs)	rho = 0.857 (n = 34)
PFree‐SAL vs. PTotal‐VEN (luteal; outlier excluded)	rho = 0.70 (n = 33)
PFree‐SAL vs. PTotal‐VEN (all follicular‐phase pairs)	rho = 0.18 (n = 29)
PFree‐SAL vs. PTotal‐VEN (follicular; outliers excluded)	rho = 0.70 (n = 21)
Source: Lu et al. (1997)		Lu‐Chicago
Luteal PFree‐SAL vs. PTotal‐VEN		r = 0.75 (n = 48), p < 0.001
Data source: Chatterton et al. (2006); samples assayed at Northwestern University	NW‐Bolivia	NW‐Chicago
PFree‐SAL vs. PTotal‐VEN (all pairs, presumed luteal)	r = 0.45 (n = 25), p = 0.026	r = 0.17 (n = 15), p = 0.55
PFree‐SAL vs. PTotal‐VEN (outliers excluded)	r = 0.43 (n = 24), p = 0.037	r = 0.48 (n = 13), p = 0.10

Figure 6 is a scatterplot of UF_LUT (y‐axis) versus UF_FOLL (x‐axis) for the 17 women in our Bolivia sample who each had both a follicular and a luteal serum/saliva pair with no evidence of contamination in either pair. The correlation between an individual's UF_FOLL and UF_LUT was low (rho = −0.19) and nonsignificant (p = 0.462). In other words, it was not the case that an individual having a relatively high follicular‐phase UF_FOLL tended to also have a relatively high luteal‐phase UF_LUT, or vice versa. This observation refutes Hypothesis 3 (i.e., that within a population, UF is consistently higher [or lower] in some individuals than in most others).

FIGURE 6.

UF_LUT versus UF_FOLL in this study of Bolivian women. N = 17 women, each having a follicular and a luteal serum/saliva pair (nine outlier P_Free‐SAL–P_Total‐VEN pairs with UF > 18% were excluded).

Figure 7 depicts the best‐fit mixed models for P_Free‐SAL as a function of P_Total‐VEN. ⁷ The model parameters changed only slightly when the nine outlier P_Free‐SAL–P_Total‐VEN pairs with UF > 18% were excluded from the analysis. Variables for cycle phase, cycle day, and reverse cycle day were not significant when included in the model and did not improve the fit of the model (analyses not shown).

FIGURE 7.

Best‐fit mixed models for P_Free‐SAL as a function of P_Total‐SER. Individual intercepts are treated as random effects. Solid line: N = 66 P_Free‐SAL–P_Total‐VEN pairs, intercept = 77.4 pmol/L (p < 0.001), β = 0.0191 (p < 0.001). Dashed line (nine pairs having a likely contaminated sample are excluded): N = 57 pairs, intercept = 121.3 (p < 0.001), β = 16.5 (p < 0.001). Insert in upper left is enlargement of boxed section of plot in lower left quadrant; blue triangles = follicular phase, red circles = luteal phase, green cross = phase uncertain; circled symbols are P_Free‐SAL–P_Total‐VEN pairs with implausibly high UF, suggesting sample contamination.

4. Discussion

Analyses of these new data from Bolivian women support the utility of, and highlight the necessary conditions for, the use of P_Free‐SAL as a biomarker of ovarian functioning and as a proxy for P_Total‐VEN in individuals and across populations.

In this set of paired saliva and venous blood (specifically, serum) samples, mean and SD for follicular‐ and luteal‐phase P_Total‐VEN and P_Free‐SAL (Table 1) are comparable to published serum reference values (Stricker et al. 2006) and “expected normal ranges” for the salivary assay (Salimetrics 2010), respectively. The similarity of the samples' CV (Table 1) for P_Total‐VEN and P_Free‐SAL in each cycle phase (respectively, 0.53 and 0.51 in the follicular phase, and 0.56 and 0.62 in the luteal phase) indicates that phase‐specific P_Free‐SAL is not substantially more or less variable than is P_Total‐VEN for the same phase. The correlations (Spearman rho) of P_Free‐SAL and P_Total‐VEN for all 66 pairs (rho = 0.858) and for the luteal‐phase pairs (rho = 0.857) are high. The correlation for follicular‐phase pairs is 0.70 once outliers with an implausible UF (≥ 19%) are excluded.

At least some of the differences in previously reported correlations of P_Free‐SAL and P_Total‐VEN arise from study differences in time lags between the collections of blood and a paired saliva sample. For this reason, in the present study in which the specific goal was to evaluate the concurrent UF, we took particular care to collect saliva near‐concurrent with the blood draw. Illustrative of this point, the two saliva–serum pairs in our study which were not collected concurrently (and were not included in the statistical analyses) had UFs of 26% and 12%, both substantially higher than the median UF in our study.

Follicular‐phase P_Free‐SAL and P_Total‐VEN pairs correlated well (rho = 0.70) once biologically implausible outliers were excluded from the analysis (circled symbols in Figure 5). Such outliers are credibly explained by saliva contamination from minute cross‐reactants (e.g., food particles, drink residues, chewed substances, ointments, make‐up [Núñez‐de la Mora et al. 2007; Vitzthum et al. 1993; Wang and Knyba 1985]). Not surprisingly, when P_Total‐VEN is very low, as is typical during the follicular phase, the impact of a contaminant is proportionally far greater than when P_Total‐VEN is high.

For example (Table 3), suppose that the true UF_Foll = 8% and UF_Lut = 2%, and that some saliva samples have a contaminant that cross‐reacts in the assay to produce a false P_Free‐SAL reading of 120 pmol/L. Then for a true mid‐follicular P_Total‐VEN of 900 pmol/L and P_Free‐SAL of 72 pmol/L, the sum of the true and false readings yields a measured P_Free‐SAL of 192 pmol/L and a measured UF of 21.3% (tall vertical black arrow in Figure 5). In contrast, for a true mid‐luteal P_Total‐VEN of 30 nmol/L and P_Free‐SAL of 600 pmol/L, adding the false reading from contamination yields a measured P_Free‐SAL of 720 pmol/L, which only raises the measured UF to 2.4% (short vertical black arrow in Figure 5).

TABLE 3.

Different impacts of same hypothetical saliva contaminant (which cross‐reacts in the assay to produce a false P_Free‐SAL reading) in follicular versus luteal phases.

Mid‐follicular	PTotal‐VEN (nmol/L)	PFree‐SAL (pmol/L)	UFFOLL (%)
True	0.9	72	8.0
Contaminant		120
Total (measured)		192	21.3

Mid‐luteal	PTotal‐VEN (nmol/L)	PFree‐SAL (pmol/L)	UFLUT (%)
True	30	600	2
Contaminant		120
Total (measured)		720	2.4

Such low‐level contamination in some samples may explain our observation that, independent of phase, P_Free‐SAL and P_Total‐VEN are highly correlated in pairs in which P_Total‐VEN > 4 nmol/L (rho = 0.807, p < 0.001), poorly correlated in pairs in which P_Total‐VEN < 4 nmol/L (rho = 0.168, p = 0.306), but moderately well correlated in pairs in which P_Total‐VEN < 4 nmol/L once samples with anomalously high UF (> 15%, evidence of likely contamination) are excluded (rho = 0.697, p < 0.001).

4.1. Hypothesis 1: UF Differs by Ovarian Cycle Phase

These new data from Bolivian women support Hypothesis 1. Excluding outliers, median UF_FOLL (=7.3%) is more than three‐fold higher than median UF_LUT (=2.2%). Previous studies have reported phase‐associated differences in mean UF of similar (Appendix A: Studies 2, 5, 7, 12) (Choe et al. 1983; Evans 1986; Tallon et al. 1984; S. Walker et al. 1981) and even greater magnitude (Appendix A: Studies 9, 13, 14) (Bourque et al. 1986; Cedard et al. 1984; De Boever et al. 1986).

To illustrate, we plotted our Bolivian data shown in Figure 7 on a log scale (Figure 8A) and data from a New Zealand (NZ) sample (Evans 1986) on the same log scale (Figure 8B), and also depicted lines of equal UF (dotted diagonals). In the NZ sample (outliers were excluded by Evans), median UF_FOLL = 2.7% (n = 10) and was three‐fold higher than UF_LUT (=0.9%, n = 23), a ratio similar to that observed in our Bolivian sample.

Also note the similar shape in the distributions of Bolivian and NZ data. In both the Bolivian and NZ samples, luteal P_Free‐SAL–P_Total‐VEN pairs (red dots) tend to monotonically increase together, hence falling within a relatively narrow band of UF values at the highest serum values. The linear regressions (red curves plotted in Figure 8A,B) of P_Free‐SAL on P_Total‐VEN for luteal phase pairs (outliers excluded) in these study samples are also similar. Furthermore, in both the Bolivian and NZ samples, the distribution of the follicular P_Free‐SAL–P_Total‐VEN pairs (blue triangles) is more globular and spans a wider range of UF values than the distribution of the luteal pairs (red dots). The distribution of data (panel 8C) from a UK sample (Wang and Knyba 1985) also resembles those in the Bolivia and NZ samples (i.e., a narrower band of UF at higher P_Total‐VEN, fanning into a wider range of UF as P_Total‐VEN decreases) with the caveat that cycle phase was not identified in the UK data (samples were collected at random during the cycle).

The sources of observed phase differences in UF are not immediately obvious. The disproportionate effect of a contaminant when P_Free‐SAL concentration is low is one factor (discussed above). However, this sort of random error is unlikely to produce the fairly consistent UF_FOLL values that have been reported by many independent studies (Appendix A). Therefore, idiosyncratic effects aside, there is presumably a biological explanation for relatively higher UF_FOLL compared to UF_LUT.

Explicitly, this difference in UF is likely a consequence of phase‐specific physiology (De Geyter et al. 2002; Hamidovic et al. 2020; Hodyl et al. 2020; Vitzthum et al. 2006). For example, Misao et al. (1999) reported that CBG mRNA level is positively correlated with P_Total‐VEN (r = 0.85, p < 0.01) and is highest at the mid‐luteal phase. Cameron et al. (2010) demonstrated that CBG affinity for P is temperature dependent over a clinically relevant range of human body temperature. They also suggested that because CBG binds preferentially to cortisol over progesterone, changes in plasma cortisol levels could result in changes in the free versus bound fractions of venous P and hence the ratio of P_Free‐VEN to P_Total‐VEN and P_Free‐SAL to P_Total‐VEN (i.e., the UF). Lewis and Elder (2011, 2013) challenged standing assumptions regarding the dynamic relationship between total CBG and cortisol (and by extension, P) with evidence that intact and elastase‐cleaved CBG may coexist in circulation.

Collectively, these studies are consistent with the hypothesis that the variation in UF with cycle phase reflects corresponding variation in the free versus bound fractions of venous P, itself related to fluctuations in CBG conformers and/or bound/free cortisol concentrations. In other words, it appears that with ovulation, there follows a progressive increase in P and upregulation of CBG mRNA, such that a lower fraction of venous P is free (and available to diffuse into saliva); but the net effect is higher P_Free‐VEN in the luteal than in the follicular phase (as was observed by Choe et al. 1983; Evans 1986; Wang and Knyba 1985; Appendix A: Studies 5, 10, 12).

4.2. Hypothesis 2: UF in Bolivian Women Differs From That of Non‐Bolivian Women

Our new data from Bolivian women refute Hypothesis 2. Specifically, the UFs in our study sample of Bolivians (Table 1, Figures 5 and 9) are consistent with other reported values for various populations (Appendix A: Studies 5, 7, 14 [Choe et al. 1983; De Boever et al. 1986; Tallon et al. 1984]) but differ from those of Chatterton et al. (2006), who argued that steroid biochemistry is somehow unusual in Bolivians. Further examination of the data used by Chatterton et al. (2006), described below, suggests more prosaic interpretations of those data.

Based on pairs of P_Free‐SAL and P_Total‐VEN samples collected at the presumptive mid‐luteal phase, Chatterton et al. (2006) reported that the mean P_Total‐VEN for NW‐Bolivia was 197% of the mean P_Total‐VEN for NW‐Chicago, but the mean P_Free‐SAL for NW‐Bolivia was only 48% of the mean P_Free‐SAL in NW‐Chicago (Table 4). This computational approach is unique to their study and therefore cannot be compared to the work done by other investigators (Appendix A). More importantly, their approach is useful (i.e., arithmetically and biologically meaningful) only if the UF is constant for all saliva–venous pairs in each of the analytical samples (i.e., only if the UF does not vary between saliva–venous pairs in the NW‐Chicago sample, and likewise for the NW‐Bolivia sample). It is evident in Figures 8D and 9C,D, that in both the NW‐Chicago and NW‐Bolivia samples there is marked variation in UF across the paired saliva–venous samples even though all pairs were collected during the presumptive mid‐luteal phase. Furthermore, the variation is greater in the NW‐Chicago sample (the two sample pairs we designated as outliers in Figure 8D were included in the analyses in Chatterton et al. (2006)). In sum, their calculation of the ratio of the group means for each analyte for the two population samples cannot address whether and to what extent the UF of paired saliva and venous samples might differ between populations.

TABLE 4.

Computational approaches for evaluating the relationship between P_Free‐SAL and P_Total‐VEN in NW‐Bolivia and NW‐Chicago samples.

Population ratio for each analyte (reported by Chatterton et al. 2006)
(NW‐Bolivia mean PTotal‐VEN)/(NW‐Chicago mean PTotal‐VEN)	30.2/15.3 = 197%
(NW‐Bolivia mean PFree‐SAL)/(NW‐Chicago mean PFree‐SAL)	252/522 = 48%

UF computation a	NW‐Bolivia	NW‐Chicago
All data (reported by Chatterton et al. 2006)	n: PFree‐SAL = 25, PTotal‐VEN = 26	n: PFree‐SAL = 17, PTotal‐VEN = 18
(A‐mean PFree‐SAL)/(A‐mean PTotal‐VEN)	0.83%	3.4%

Alternative approaches/estimates for same data used by Chatterton et al. (2006)
Only paired data (PFree‐SAL & PTotal‐VEN)	n pairs = 25	n pairs = 15
(A‐mean PFree‐SAL)/(A‐mean PTotal‐VEN)	0.87%	3.1%
A‐mean (individual PFree‐SAL/PTotal‐VEN)	1.97%	12.1%
(G‐mean PFree‐SAL)/(G‐mean PTotal‐VEN)	1.17%	5.0%
G‐mean (individual PFree‐SAL/PTotal‐VEN)
Median (individual PFree‐SAL/PTotal‐VEN)	0.74%	4.4%
Paired data without outliers	n pairs = 24	n pairs = 13
(A‐mean PFree‐SAL)/(A‐mean PTotal‐VEN)	0.88%	2.5%
A‐mean (individual PFree‐SAL/PTotal‐VEN)	2.04%	5.1%
(G‐mean PFree‐SAL)/(G‐mean PTotal‐VEN)	1.2%	3.45%
G‐mean (individual PFree‐SAL/PTotal‐VEN)
Median (individual PFree‐SAL/PTotal‐VEN)	0.77	3.14%

^a A‐mean = arithmetic mean; G‐mean = geometric mean.

Chatterton et al. (2006) also reported (A‐mean P_Free‐SAL)/(A‐mean P_Total‐VEN) using all measurements, regardless of whether they were paired (Table 4). This approach appears to be the most frequently used by others (Appendix A); however, it is often not clear in these studies whether a previously reported UF included all measurements or only paired data. Using non‐paired data means that any between‐day or between‐individual variation in P_Free‐SAL, P_Total‐VEN, and/or UF within a group confounds the inter‐group UF difference (see discussion in Section 2.4 and Appendix B). Calculated using all data, UF_Lut for NW‐Bolivia = 0.83%, similar to the range of values reported by others for the luteal phase (Appendix A), and NW‐Chicago = 3.4%, greater than any of these previous reports. Table 4 also presents UF for each NW sample using alternative computational approaches and only paired data. Notably, for NW‐Chicago the arithmetic mean of all the individual UFs is 12.1%, an exceptionally high mean UF that strongly suggests the presence of one or more outliers and/or other anomalies in the NW‐Chicago data.

The radioimmunoassay (RIA) protocol used by Chatterton et al. (2006) was first published by Lu et al. (1997), who reported r = 0.75 for paired P_Free‐SAL–P_Total‐VEN samples collected during the luteal phase from 48 Chicago women (designated here as “Lu‐Chicago”). In contrast, the correlation is only r = 0.17 for NW‐Chicago (n = 15) and r = 0.45 for NW‐Bolivia samples (n = 25) (Table 2). Excluding three extreme outliers (which lower r and strongly bias the group means), the correlation is still much lower in both samples (NW‐Chicago r = 0.48; NW‐Bolivia r = 0.43) than those reported by Lu et al. (1997) and nearly all other researchers (Appendix A). These much lower correlations strongly suggest—particularly in the case of the NW‐Chicago sample—that these assay data do not accurately represent the true hormone levels in the study participants, most likely due to failed laboratory protocols and/or sample contamination or degradation.

Some of the problems with the NW‐Chicago and NW‐Bolivia data are evident in the scatterplot of the data and the plotted regression line fitted to the non‐outlier luteal‐phase data in each sample (Figure 8D). In such a scatterplot, an ideal data set (P_Free‐SAL highly correlated with P_Total‐VEN) would have all the points representing luteal‐phase pairs falling on a relatively narrow diagonal band of UF and those points representing follicular‐phase pairs falling within a somewhat broader band with biologically reasonable values and ranges for each hormone and each UF. Correspondingly, the luteal‐phase regression model would have an intercept that is only a tiny fraction of a typical mid‐luteal P_Free‐SAL, so that the linear (“slope”) term in the model contributes a large majority of the modeled P_Free‐SAL for almost all luteal data. Such patterns are seen in Figure 8A–C. The NW‐Chicago and NW‐Bolivia data are far from this expectation. In particular, there are extreme outliers. Two of the NW‐Chicago P_Free‐SAL values are excessively high (> 900 pmol/L), and one NW‐Bolivia pair (boxed red dot in Figure 8D) has the implausible combination of a very low P_Free‐SAL (51 pmol/L), suggesting an anovulatory cycle or a follicular sample, paired with a moderately high P_Total‐VEN (13.7 nmol/L), typical of an ovulatory cycle's luteal phase. These outliers severely bias the sample means, rendering questionable any interpretations based on the means reported by Chatterton et al. (2006) (see Table 4).

Moreover, the data for the NW‐Chicago sample are actually more anomalous than those for the NW‐Bolivia sample and thus are not explained by the hypothesis that Bolivians have unusual physiology and/or biochemistry. Figure 9 presents histograms and cumulative distributions of UF by cycle phase (red = luteal, blue = follicular) in Bolivia (this study, panel 9A) and NZ (panel 9B), and for mid‐luteal phase sample pairs in NW‐Chicago (panel 9C) and NW‐Bolivia (panel 9D); the x‐axis is log scale UF. In both Bolivia and NZ, the median luteal phase UF is lower and the variability is less than for the respective follicular‐phase UF. The luteal phase UF is much more variable (shown in Figure 9 as the distributions being much broader and the IQR “error bars” being much wider) in both NW samples (Figure 9C,D) than in the Bolivia (Figure 9A) or NZ (Figure 9B) samples. In fact, the NW‐Chicago sample has a more variable luteal UF (IQR = 3.45%) than the NW‐Bolivia sample (IQR = 1.63%).

The effects of these anomalies are evident in Figure 8D. Most notably, in both the NW‐Chicago and NW‐Bolivia samples (each comprising only luteal phase measurements), P_Free‐SAL shows little systematic variation even as P_Total‐VEN ranges over more than a factor of 20. UF varies over almost as wide a range as P_Total‐VEN but in the opposite direction to P_Total‐VEN so that their product (P_Free‐SAL) varies only weakly. This pattern is captured by the plotted regressions for NW‐Chicago (black line) and for NW‐Bolivia (red line), both of which are notably flatter (i.e., have much higher intercepts relative to the mid‐luteal P_Free‐SAL) than those in NZ (Figure 8B) and Bolivia (Figure 8A). This relative constancy of P_Free‐SAL across the luteal phase in both of the NW study samples is a biologically inexplicable pattern that has never been previously reported. In contrast, other published studies generally show that for data restricted to the luteal phase, P_Free‐SAL is roughly proportional to P_Total‐VEN with a relatively narrow range of UF (see, e.g., Figure 9A,B).

We find sample contamination and/or inaccurate assays to be a more parsimonious explanation for the results reported by Chatterton et al. (2006). They suggested possible differences in bioavailability in the NW‐Bolivia sample as an explanation for their findings. However, their results are actually more anomalous (little systematic variation in P_Free‐SAL while P_Total‐VEN varies over a wide range) for the NW‐Chicago sample than for the NW‐Bolivia sample (Figure 8D).

Moreover, if UF actually varied in the manner suggested by their presented data, then it would be difficult to explain how many other published studies of several different populations (Ellison et al. 1989; Jasienska and Ellison 1998; Panter‐Brick et al. 1993), including Bolivians (Vitzthum 2013; Vitzthum et al. 2002, 2004) and Chicagoans (Lu et al. 1997), have obtained reasonable profiles of P_Free‐SAL during the course of an ovarian cycle (Figure 10).

FIGURE 10.

Salivary progesterone profiles of samples from four populations, measured in the same laboratory using the same assay protocol (Ellison et al. 1993).

4.3. Hypothesis 3: UF is Consistently Higher (Or Lower) in Some Individuals Than in Most Others

Hypothesis 3, proposed by Konishi et al. (2012), was not supported by our analyses of new data from Bolivians. We found no evidence of subsamples of women that had consistently higher or lower UF than the sample average. Notably, an individual's follicular and luteal phase UFs were not significantly correlated (Figure 6). None of the nine Bolivian women exhibiting an unusually high UF (> 19%) for one saliva–serum pair exhibited a particularly high UF in her other saliva–serum pair. Notably, P_Total‐VEN was < 1.0 nmol/L in all nine cases in which UF > 19%. The most parsimonious explanation for these nine outliers is undetected contamination in samples of particularly low P_Free‐SAL (i.e., low signal to noise ratio as in the hypothetical example presented in Table 3). Consistent with this explanation, UF was 12%–59% in samples in which subjects had visible gum bleeding (these corrupted measures were neither plotted nor included in our analyses).

Konishi et al. (2012) evaluated P_Free‐SAL (measured using Salimetrics EIA kits) and P_Total‐VEN (measured in DBS collected from a pricked fingertip) in samples of self‐identified Caucasian or Japanese‐ancestry women living in the Seattle area. Based on standard (fixed effects) hierarchical linear regression (pooled sample of 232 observations, 4 weekly paired saliva‐DBS samples from each of 58 women), they observed that the UF in some women were either consistently above or below the regression line, attributed such patterns to “individual‐specific characteristics perhaps salivary hormone excretion capacity,” and concluded that “salivary P4 [does] not closely mirror between‐individual variation of serum P4…in a substantial proportion of individuals.”

The differences in findings between Konishi et al. (2012) and our study (and also the past studies presented in Appendix A) likely arise from their analytical approaches. In a repeated measures study design (i.e., any study design where samples are collected at multiple times from the same participant), the samples from a single individual are not independent, thus precluding the use of standard linear regression. Mixed‐effects regression models, which we used to evaluate our Bolivian data, can address this nonindependence (Twisk 2006; West et al. 2007) and can (should) incorporate per‐individual random effects to model inter‐individual variation in overall hormone levels. Twisk emphasized that the inclusion of a random intercept is a conceptual necessity in regression models of repeated observations.

Konishi et al. (2012) treated 232 paired samples as independent observations in a standard linear regression analysis. Women with four positive residuals were designated as “high secretors,” those with four negative residuals were designated as “low secretors,” and those with a mix of positive and negative residuals were designated as “medium secretors.” However, the nonindependence of the four sample pairs from each woman means that a woman with one positive residual is more likely to have four positive residuals than would be predicted by random chance alone. Although Konishi et al. (2012) mentioned mixed‐effects models and included the mixed models' coefficients in their table 3, their interpretations and conclusions appear to be based solely on the fixed‐effects models.

In addition, the distribution of their P_Free‐SAL data at a given P_Total‐DBS is quite skewed (has a long right tail; see their figure 3A). This means that, contrary to the assumption of their calculation of the expected number of high and low secretors in their sample, each individual residual is not expected to have an equal (50%) probability of being positive or negative. Instead, we expect > 50% negative residuals (relatively smaller in magnitude) and < 50% positive residuals (relatively larger in magnitude). This expectation is broadly consistent with their finding that 40% (23/58) of individuals had four negative residuals, but only 20% (11/58) had four positive residuals. A more precise assessment is not possible due to the nonindependence (correlation) of the four residuals from a single woman.

Another factor that may have confounded Konishi et al.'s analyses is their stated disregard of individual differences in cycle length or whether a cycle was ovulatory (even though they did collect these data). Unfortunately, their approach is likely to upwardly bias the number of follicular‐phase saliva–blood pairs given their sample collection schedule. The paired samples for their analyses were collected once per week for 4 weeks from each of 48 women, but it is unclear whether the collections were all on the same cycle days. Even if that were the case, those women with longer follicular phases (which is more variable in length than the luteal phase) may have contributed three follicular‐phase samples and only one luteal‐phase sample. An additional 10 women collected daily samples at home; of these, cycle days 3, 10, 17, and 24 were included in the analyses (for a total of 58 cycles of which 17 were missing a cycle end date). Of course, anovulatory cycles, by definition, would not have contributed any luteal phase samples (2 cycles > 73 days length were considered anovulatory but retained in the analyses; as no other criteria for anovulation were considered, all other cycles, some of which may have been anovulatory, were also retained).

As discussed earlier, UF differs by cycle phase in ovulatory cycles, and the low P_Total‐VEN values in the follicular phase are more likely to have lower signal to noise ratios. Nonetheless, Konishi et al. (2012) (p. 239) judged anovulation to be “irrelevant…because statistical analyses were conducted without taking cycle phase into account.”

As Konishi et al. (2012) did not consider these various sources of potential bias, interpretation of their findings is difficult. The specific cause of the greater similarity among the samples and residuals from a single woman cannot be ascertained from the provided information. Rather than positing unknown mechanisms, there are several more parsimonious explanations for their reported observations, including:

• a participant's personal sample collection practices (e.g., a participant who typically shortens the waiting period after rinsing may have consistently lower measured UF than would otherwise be the case, and one who has undetected gum bleeding and/or poor tooth brushing/flossing technique, leaving debris in the mouth, may have consistently higher measured UF),

• the use of DBS from finger pricks rather than serum from venipuncture,

• delays between collecting saliva and DBS,

• delayed or inadequate freezing of sample,

• inclusion of four samples from each anovulatory cycle, all of which can be expected to have low progesterone and be more susceptible to contamination (which could produce a falsely elevated UF),

• other plausible biological factors (e.g., variation in CBG and/or cortisol concentrations).

In sum, it is not surprising that more women have all positive or all negative residuals in Konishi et al. (2012)'s regression analyses than would be expected for independent random variation of residuals across individuals and samples (each with an independent 50%/50% probability of being positive/negative), and it cannot be inferred that such a pattern is likely to be the consequence of hypothesized individual “secretor capacity.” As discussed earlier, P_Free‐SAL enters saliva via passive diffusion, the mechanics of which have not been shown to vary significantly between individuals. Studies of saliva flow rate and of salivary metabolism have reported neither factor to be an important source of error or variation in P_Free‐SAL (Blom et al. 1993; Laine and Ojanotko 1999).

4.4. Advantages and Limitations of This Study

Our sample collection protocol was specifically optimized for high‐quality saliva collection (see Section 2.2). Advantages of this study include:

1. Pre‐collection individual training for each study participant included demonstration and practice of saliva collection protocol.

2. Participants abstained from breakfast the morning of sample collection to reduce the possibility of food‐particle contamination of saliva samples.

3. A research team member monitored saliva collection to ensure compliance with protocol.

4. Blood sampling followed saliva collection to minimize any negative impact of blood draws on saliva sample (passage of P_Free from blood to saliva is only about 1 min [Blom et al. 1993]).

5. Follow‐up with participant to learn first day of subsequent menses (used to calculate reverse cycle day and estimate day of ovulation).

6. All sampling in the morning (before lunch) to minimize the potential effect of circadian variation.

7. Conducted phase‐specific analyses.

8. Used commercial EIA kits. Most studies that have examined the relationship between P_Free‐SAL and P_Total‐VEN have used RIA assays (Table 1). EIA dispenses with the need to handle and dispose of radioactive materials and is now more commonly used in many research projects and smaller labs. But fewer studies have used EIA (instead of RIA) to assess the correlation between P_Free‐SAL and P_Total‐VEN, a gap this study narrows.

9. Use of commercial assay kits makes it easier for other researchers to replicate/generalize our results.

Limitations of this study include:

1. No direct measure of occurrence and timing of ovulation.

2. Only two paired saliva/blood collections (follicular phase and presumptive luteal phase) per woman.

3. Because our study participants met selection criteria (e.g., regularly cycling and not using hormones) intended to minimize the effects of potential confounders, the study sample is not strictly representative of the population of Andean Bolivian women. In the absence of modern contraception, fertility is high (≥ 7 children/family) in Andean Bolivian women (Stinson 1982). The sample's low parity is not surprising (half of the sample are university students) but contrasts with reproductive patterns in the larger population, especially those in rural Andean communities where there is little access to contraception. Although early investigators had suggested high altitude might impair successful reproduction, this hypothesis was not supported by subsequent demographic data (reviewed in Vitzthum (2013)). Notably, there is evidence for natural selection for adaptation to hypoxic environments in Tibetan populations (Ye et al. 2024), but comparable work has yet to be undertaken in Andean populations.

4. Collectively, the study participants were a convenience (rather than random) sample and likely were, on average, economically and socially better off than the average Andean Bolivian woman. The participants included residents from a poorer neighborhood (where we had conducted a previous study [Vitzthum et al. 2002]). Some were maids/housekeepers or had other low‐paying and/or occasional jobs. The economic status of the study participants attending university was more varied, roughly ranging from low to high middle class; it is unlikely any students were very poor or especially well‐off.

5. Summary and Broader Implications

UF in this study sample of Bolivian women did not differ in any substantial or consistent manner from UF observed in women in other populations. In addition, an individual's follicular‐ and luteal‐phase UF were not correlated, refuting the hypothesis that there are Bolivian women who consistently have relatively higher or lower UF than the sample's mean UF. Mean UF did vary by cycle phase, being generally higher and more variable before ovulation than after. Prior research on the biochemistry of progesterone transport has consistently found that progesterone UF varies between cycle phases. To the best of our knowledge, there is no credible evidence that phase‐specific UF of progesterone substantially differs between human populations or individuals. However, this possibility has not yet been thoroughly investigated. The sources of phase‐associated variation in UF deserve additional study, particularly regarding temporal and individual variation in P_Total‐VEN and P_Free‐SAL, and their dynamic relationship to variation in total and different conformers of CBG, and perhaps also to variation in free and bound cortisol.

There is substantial evidence of natural variation in progesterone concentration between populations and between healthy individuals within populations (Figures 1, 3, and 10). At the same time, variability (measured by the ratio of the 95th to the 5th percentile) in progesterone biomarkers is similar in samples of Germans (mean‐peak‐pregnanediol‐glucuronide (PdG) varies 3.65‐fold) and Bolivians (mean‐peak‐P_Free‐SAL varies 3.9‐fold) (Vitzthum et al. 2021). Although there are several hypotheses, the cause(s) of progesterone variation are uncertain. There is, nonetheless, clear empirical evidence that Bolivian women successfully reproduce (averaging about 7–8 pregnancies in a lifetime) with an average progesterone concentration only 70% that of US women (Vitzthum et al. 2004). These and other findings raise questions regarding the criteria for defining “normal” progesterone levels.

All too often, it is assumed without even a semblance of justification that the observed biology of Western Europeans and their descendants (aka “white”) is “normal” and thus the natural yardstick by which to assess the normalness of a phenotype in other populations. With regard to menstrual cycling, this myopia has been repeatedly interrogated (Eaton et al. 1994; Strassmann 1997; Vitzthum et al. 2021) and yet it persists, as witnessed by the assumption that the progesterone data from the NW‐Chicago sample were assumed to be correct and, therefore, those from the NW‐Bolivia sample were presumed to be aberrant. To the contrary, compared to the findings from other studies, it is the NW‐Chicago data that were decidedly more “abnormal.”

General acknowledgment of the extensive natural (i.e., healthy and “normal”) variability in ovarian cycling, between individuals and populations, has been slow in coming despite having been extensively documented (Arey 1939; Foster 1889; Fraenkel 1926; Treloar et al. 1967). Because the timing of hormonal changes during the ovarian cycle is uncertain, a single measurement of a cycle hormone provides limited information regarding ovarian functioning. The common practice of collecting data only during the early follicular phase in order to avoid accounting for the effects of ovarian steroid variation has hampered basic research and clinical trials in menstruating persons. The serial collection of salivary biomarkers, including P_Free‐SAL, has yielded otherwise unattainable insights regarding variation in human reproductive functioning. There are, however, many aspects of saliva collection, storage, shipping, and processing that require consistent adherence (Ellison 1988; Gröschl 2017; Vitzthum 2021). Appendix C discusses some of the more notable challenges in the assessment of ovarian functioning. Neglecting to address these issues can lead to spurious outcomes and misinterpretations of the data.

With due attention to the collection and handling of saliva samples, P_Free‐SAL is a suitable biomarker for evaluating temporal, individual, and population variation in P_Total, particularly if repeated sampling is done during the luteal phase when P_Total is elevated in ovulatory cycles. Investigators have often cautioned that a single P measurement has only limited research or diagnostic utility. A principal advantage of P_Free‐SAL is serial sample collection where such would be difficult or impossible with venous blood collection. In novel contexts, a selected biomarker should be pre‐tested to evaluate research team and study participant compliance to protocols and setting‐ and culturally‐specific challenges (including logistics and unanticipated sources of sample contamination). Data on likely and potential covariates of hormonal variation (e.g., age, ethnicity, socioeconomic resources) should be collected. As did Wang and Knyba (1985), it is very valuable to future investigators to publish all data and measurements, even for those cases where the sample measurements were below assay sensitivity, and to calculate sample statistics that both include and omit such measurements and other outliers.

Often, a biomarker is a surrogate or approximation of the biological status or process of interest, but is not of interest in and of itself. Careful protocols can reduce error in the biomarker, but less error does not necessarily make the biomarker a more suitable proxy for the focus of interest. The history of the use of body mass index (BMI) exemplifies this distinction. There has been extensive effort to make measurements of height and weight more precise, and by extension the calculation of BMI more accurate. Nonetheless, only relatively recently has there been a comprehensive reconsideration of whether BMI is a suitable indicator of body composition in all populations (Heymsfield et al. 2016).

Selection of a biomarker should be based on the research question, the research context, and the available resources. Because confidence in the selected biomarker is requisite to a successful study, the suitability of biomarker options should be evaluated prior to initiating the bulk of sample collection. Different steroids have different properties, functions, and carrier molecules. The biochemistry of progesterone cannot be generalized to all steroids or even all reproductive steroids.

Gender‐based health disparities are pervasive throughout the world regardless of an individual's or population's social and economic status. Women's health is manifestly understudied, a reflection of both cultural and political views of female bodies and of the challenges of adequately accounting for hormonal changes during the menstrual cycle. Over‐representation of data collected during the early follicular phase does not suffice. We can and must do better. Some necessary steps to improve research and health care are obvious and relatively easy (adequate funding to collect data throughout menstrual cycling; development, monitoring and improvement of tools and protocols for measuring hormonal variability). Other steps are more challenging (accounting for the dynamic interactions between the biological and social determinants of health).

Just as it is costly to continue our work with tools not up to the task, so it is equally costly to discard useful tools without good reason. The development (and improvement through replication) of a robust toolkit for assessing changes in and the impacts of menstrual cycle hormones is foundational to reducing gender‐based health disparities.

Author Contributions

All coauthors contributed to the study execution and reviewed the manuscript for intellectual content. V.J.V. and D.B. conceived and supervised the study. V.J.V., L.E., and E.C. conducted data collection in Bolivia. V.J.V. and J.T. performed statistical analyses and wrote the manuscript.

Ethics Statement

All study protocols were approved by the institutional review boards at the Medical School of Universidad Mayor de San Andrés, La Paz, Plurinational State of Bolivia, and Indiana University, Bloomington, USA.

Consent

All study participants gave signed informed consent.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1. Supporting information.

Appendix A. Assessments of the Correlation Between Progesterone in Blood and Saliva Samples

Author and year	Population/location	n persons (n paired samples) (L: luteal; F: follicular)	r (mean [range]) (L: luteal)	p	Saliva assay	Blood assay [PTotal‐VEN unless noted otherwise]	Apparent uptake fraction (UF)
							Follicular	Luteal
(1) R. F. Walker et al. (1979)	UK (United Kingdom)	4 (L: 40; 8–15/person)	L: [0.91–0.97]		RIA	RIA/plasma
(2) S. Walker et al. (1981) a , b	UK	5 (103; 14–25/person)	[0.81–0.93]		RIA	RIA/plasma
		9 (143; 8–25/person)	0.86 [0.81–0.97]		RIA	RIA/plasma	2.3%	1.2%
(3) Chearskul et al. (1982) c	UK	5 (“several” during cycle)	Parallel changes but nonlinear correlation		RIA	RIA/serum	≈ 3%	≈ 0.9%
(4) Shah and Swift (1984)	UK	50 (50; random day)	0.89		RIA	RIA/plasma
(5) Choe et al. (1983) d	US (United States)	9 ovulatory cycles (96)	0.58 (L: 0.45)	0.001 L: < 0.001	RIA	RIA/plasma	8%	1.5%
(6) Donaldson et al. (1984)	UK	2 (10; 18) L: 1 day, ~hourly			RIA	RIA/plasma		~ 1%
(7) Tallon et al. (1984) e	Ireland	11 (56)	0.85	< 0.001	EIA	EIA/plasma	7.5%	1.8%
(8) Zorn et al. (1984) f , g	France	32 (L: 80)	L: 0.76	< 0.001	RIA	RIA/plasma	5.1%	1.02% ± 0.39%
(9) Cedard et al. (1984) h	France	14 (L: 39)	L: 0.63	< 0.001	RIA	RIA/plasma	5.3%	1.12% ± 0.39%
(10) Wang and Knyba (1985)	UK	36 (36)	0.78	< 0.001	RIA	RIA/serum
		36 (36)	0.75	< 0.001	RIA	PFree‐VEN: RIA/serum
(11) Webley and Edwards (1985)	UK	(77)	0.71	< 0.001	RIA	RIA/plasma
(12) Evans (1986) i	New Zealand	27 (L: 25; F: 18)	0.90		RIA	RIA/plasma
		4 (38)	0.83		RIA	RIA/plasma	2.1%	0.87%
(13) Bourque et al. (1986) j	Belgium	14 (76)	0.68 L: 0.78 F: 0.04	< 0.001 L: < 0.01 F: ns	RIA	RIA/plasma	9.6%	1.6%
(14) De Boever et al. (1986) k	Belgium	7 (96)	0.88	< 0.001	CIA	RIA/serum	Early‐mid F: 19.6% Peri‐ovul: 6.3%	1.8%
(15) Heasley and Thompson (1986)	Ireland	40	0.82	< 0.001	RIA	RIA/serum
(16) Petsos et al. (1986)	UK	6 (82)	0.89 [0.78–0.97]	< 0.002	RIA	RIA/serum
(17) Vuorento et al. (1989)	Finland	32 (40)	0.93	N/A	RIA	RIA/serum
(18) Wong et al. (1990)	People's Republic of China	10 (43)	0.93	< 0.001	RIA	RIA/serum
(19) Kesner et al. (1992)	US	10 (7–15 pairs/person; ~2nd & 3rd cycle weeks)	0.73 (mean r of 10 individual r)	≤ 0.0001	RIA	RIA/serum
(20) Lipson and Ellison (1992)	US	3 (“over course of menstrual cycle”)	0.80–0.97		RIA	RIA/serum
(21) Bolaji (1994)	Ireland	23	0.99	< 0.0001	EIA	EIA/serum
(22) Vienravi et al. (1994)	Thailand	20 (~40)	0.835	< 0.001	RIA	PFree‐VEN : RIA/serum
(23) Chearskul and Visutakul (1994) l	Thailand	10 ovulatory cycles (daily: 280 pairs) (L: 138; F: 132)	All: 0.89 L: 0.75 F: −0.08	< 0.001 < 0.001 ns	RIA	RIA/plasma	3.32% ± 0.32%	1.39% ± 0.07%
(24) Lu et al. (1997)	US	48 (L: 48)	L: 0.75	< 0.001	RIA	RIA/serum
(25) Abood (2008)	Iraq (16–19 years)	9	0.73	0.02	RIA	RIA/serum
	(20–29 years)	26	0.75	0.001	RIA	RIA/serum
	(30–39 years)	42		ns	RIA	RIA/serum
	(40–49 years)	11	0.65	0.03	RIA	RIA/serum
(26) Houghton (2008) m	Britons (European descent)	10 (L: 8, F: 1, ?: 1)	0.88	< 0.0005	RIA	RIA/serum		0.30% ± 0.18%
	Bangladeshi adulthood migrants to UK	18 (L: 9, F: 1, ?: 8)	0.19	ns (0.448)	RIA	RIA/serum		1.10% ± 2.2%
	Bangladeshi sedentee	11 (L: 10, F: 1)	0.70	< 0.0005	RIA	RIA/serum		0.25% ± 0.23%

^a n = 9 comprises n = 5 reported by S. Walker et al. (1981) and n = 4 reported by R. F. Walker et al. (1979).

^b UF for n = 9 reported by Riad‐Fahmy et al. (1987): table II.

^c UF estimated from data reported by Chearskul et al. (1982): figure 8.

^d UF estimated from means reported by Choe et al. (1983).

^e UF estimated from means reported by Tallon et al. (1984).

^f Follicular UF reported by Bourque et al. (1986): table 2.

^g Luteal UF reported by Zorn et al. (1984): table 1 (calculated from individual saliva/plasma pairs).

^h Follicular UF (mean saliva P)/(mean plasma P) and luteal UF (calculated from individual saliva/plasma pairs) reported by Cedard et al. (1984).

ⁱ UF estimated from mean of individual paired ratios reported by Evans (1986).

^j UF reported by Bourque et al. (1986): table 2 (calculated as mean‐saliva/mean‐plasma).

^k UF estimated from data reported by De Boever et al. (1986).

^l UF reported by Chearskul and Visutakul (1994): table 2 (calculated from individual saliva/plasma pairs).

^m UF reported by Houghton (2008): calculated from individual saliva/plasma pairs.

Appendix B.

Rather than using computation (1), presented in Section 2.4.1, a preferable if less common way to calculate UF is to use the A‐mean of the individual UFs:

$$\mathrm{UF}\left(\mathrm{A}‐\text{mean of}\mathrm{UFs}\right)=\text{mean}\left({\mathrm{UF}}_i\text{for}i=1,2,3,\dots ,N\right)$$ (B.1)

where each individual UF is defined as UF _i  = P_Free‐SAL,i /P_{Total‐SER,i} .

Unfortunately, while both (1) and (B.1) would seem to be “natural” definitions for “mean UF,” they generally give different answers (see Table 4 in Section 4.2 for several examples), and it is not obvious whether (1) or (B.1) is “more natural.”

Another problem with both (1) and (B.1) is that when calculating the A‐mean of a set of values with a wide dynamic range, the A‐mean is relatively much more sensitive to changes in the larger values and less sensitive to changes in the smaller values. For example, given a set of samples from both the follicular and luteal phases, if we were to (hypothetically) double a single (relatively large) mid‐luteal P_Free‐SAL,i value (leaving all the other P_Free‐SAL,i and all the P_{Total‐SER,i} unchanged), computation (1) would change much more than if we were to double a single (relatively small) mid‐follicular P_Free‐SAL,i value. Similarly, if we were to (hypothetically) double a single (relatively large) mid‐follicular UF _i (leaving all the other UF _i unchanged), computation (B.1) would change much more than if we were to double a single (relatively smaller) mid‐luteal UF _i .

Because of these problems, we suggest that “mean uptake fraction” is better defined as the geometric mean (G‐mean):

(B.2)

or alternatively as

(B.3)

where the G‐mean of a set of N numbers X ₁, X ₂, X ₃, …, X _N is defined as

$$\mathrm{G}‐\text{mean}\left(X_i\text{for}i=1,2,3,\dots ,N\right)=N^{\mathrm{th}}\text{root of}\left(\left(X_1\right)\left(X_2\right)\left(X_3\right)\dots \left(X_N\right)\right)$$

Mathematically, computations (B.2) and (B.3) always give precisely the same result, so it doesn't matter which of these we choose to define “mean uptake fraction.” Moreover, the G‐mean has the same relative sensitivity to both small and large values (e.g., hypothetically doubling a single value causes the same percentage change in the G‐mean regardless of whether the doubled value is relatively small or relatively large).

Another possible definition (offering improved robustness to outliers) is to take the median of the individual uptake fractions: UF(median of UFs) = median(UF _i for i = 1, 2, 3, …, N). If the distribution of UF is non‐symmetric (as is often the case), the median UF is a more informative measure of central tendency than the A‐mean or G‐mean UF, or alternatively the UF _i values may be log‐transformed.

Appendix C. Notable Challenges in the Assessment of Ovarian Functioning

Accurately characterizing individual and population variation in ovarian steroids requires fastidious attention to sample collection and judicious evaluation of the data. Regardless of the chosen matrix and assay protocols, the inherent nature of ovarian functioning presents several potential pitfalls that should be recognized, and addressed accordingly, in nearly any measurement of ovarian hormones.

C.1. Identifying Anovulatory Cycles and Cycle Phases

Anovulatory cycles are fairly common in many populations and subpopulations, particularly those living in demanding physical and socioeconomic conditions. Reported anovulation rates in adult premenopausal non‐breastfeeding women range from only 8% to more than 50% (Vitzthum 2009). Obviously, an anovulatory cycle has no luteal phase and therefore will not have the expected rise in postovulatory steroids. Progesterone concentration and UF in the third or fourth week post‐menses onset will resemble those of a follicular phase. Furthermore, cycle length is not reliably indicative of ovulatory status, therefore reverse day counting to estimate the timing of a putative ovulation and define phases is prone to error. There are several methods for evaluating the ovulatory status of a cycle and estimating its timing, the choice of which depends on resources, research population and setting, and research questions (Vitzthum 2021).

C.2. Sampling Frequency: No Matter the Matrix, One Sample Per Cycle Likely Won't Suffice

Despite substantial evidence of marked variability in ovarian hormones between individuals and populations as well as during the menstrual cycle, the inadequacy of a single sample for many clinical and research purposes remains widely underappreciated (Ansbacher 1990; Arslan et al. 2023; Fujimoto et al. 1990). Natural variability in P arises from several factors including pulsatile hormone release and individual circadian and ultradian rhythms (Delfs et al. 1994; Filicori et al. 1984; Veldhuis et al. 1988). Factors that confound interpretation of a single measurement include minute contaminants, undetected bleeding, type of assay and its sensitivity, different reference standards and other laboratory idiosyncrasies, and insufficient adherence to collection and storage protocols, among others (Fujimoto et al. 1990).

Thus, a single sample, whatever the matrix, is likely inadequate for accurately characterizing individual and populational variation in ovarian steroids, for ascertaining whether ovulation has occurred or for testing many other questions of interest.

For example, following the advent of RIA for ovarian steroids, small anthropological studies (van der Walt et al. 1977, 1978), and several large epidemiological studies that investigated hypothesized links between breast cancer and ovarian steroid variation, collected only a single venous sample per person (reviewed in Vitzthum 2009). The sparse sampling in these early studies significantly undercut useful interpretation of the data.

As well, a high P concentration in a solo sample does not necessarily indicate the occurrence of ovulation; the sample may have been contaminated. In an ovulatory cycle, ovulation is accompanied by a continuing rise in progesterone, which is best assessed using serial sampling (i.e., at least every other day). In addition, an inordinately high progesterone reading during the follicular phase is not necessarily indicative of a contaminant. In some instances, a persistent elevation in follicular phase progesterone is associated with early pregnancy loss (Vitzthum et al. 2006), a pattern that would have been unrecognized if not for serial sampling.

A single elevated P_Free‐SAL or P_Total‐VEN measurement taken at some time in the middle of the second week following the first appearance of vaginal bleeding may occur for various other reasons including a short follicular phase, miscounting days since bleeding, a very early pregnancy loss, or fortuitously capturing a pulsatile P spike or a preovulatory P spike. A low P_Free‐SAL or P_Total‐VEN measurement from a putative mid‐luteal day may indicate a long follicular phase, anovulation, or assay failure. Serial sampling, aggregating repeated measures, and examining the data set for obvious outliers based on statistical methods and biological plausibility substantially mitigates such errors.

Wang and Knyba's (1985) evaluation of the relationship between P_Free‐SAL and P_Total‐VEN exemplifies judicious assessments of measurements and illustrates the limitations of relying on a single sample. They collected paired saliva and blood samples on a single randomly selected day from 96 volunteers in the United Kingdom. Seven of these pairs had P_Total‐VEN < 10 nmol/L (indicative of follicular phase) but P_Free‐SAL > 300 pmol/L (two were ≥ 750 pmol/L), each of which could be mistaken for luteal phase P if relied upon as the sole evaluation of menstrual functioning. Despite their rigorous efforts to prevent such contamination, they concluded (p. 978) that food debris was the most likely immunoreactive impurity because some “saliva samples would need to have up to 25% of blood present” for the observed readings and “a subsequent saliva sample from such a volunteer did not have a high progesterone level.”

Other potential sources of error in the measurement of salivary steroids (which are particularly “sticky” molecules) include the use of collection devices and plastics other than polypropylene (Banerjee et al. 1985; Wood 2009; Vitzthum 2021). Unfortunately, findings from some older studies that followed then contemporary standards are now of uncertain validity because of the materials used. There's still the occasional study that unwittingly uses unsuitable collection materials (e.g., polyethylene).

C.3. Essential Precautions for Saliva Collection and Interpretation of Steroid Measurements

Our study highlights the unavoidable necessity of giving as much attention to the collection, transport, and storage of saliva samples as is typically given to the assays of those same samples. The ease with which saliva is readily obtained in almost any setting belies the risks of contamination and degradation that may render the sample useless and generate spurious findings. There are several resources on best practices for saliva sample collection, handling, and interpretation of data (see Vitzthum (2021) and references therein). Sample handling, transport, and storage from shortly after collection until assaying must be designed to minimize bacterial growth and biomarker degradation. A preservative may be added to the sample, but some preservatives can interfere with some assays (see Chester and Vitzthum (2018) for an EIA protocol that can be used with samples preserved with sodium azide). Freezing shortly after collection is ideal but may be difficult in remote locales and requires the maintenance of a cold chain until the samples are assayed (repeated freeze/thaw cycles degrade many biomarkers).

Undetected food particles, drink residues, chewed substances (e.g., tobacco, coca leaves, betel nut), ointments, makeup, and blood can result in erroneously elevated measurements. Rinsing with cool clean water can mitigate the presence of these spoilers, but there must be enough time after rinsing to allow equilibration of normal saliva composition and avoid erroneously low measurements. It is fairly easy to ensure adherence to collection protocols in a laboratory but obviously far more difficult to do so if study participants are self‐collecting alone at home or elsewhere. For example, a participant who typically shortens the waiting period after rinsing may have consistently lower UF than would otherwise be the case, and one who has undetected gum bleeding may have consistently higher measured UF. In our previous community‐based studies in Bolivia (Vitzthum et al. 2006, 2009), each saliva sample was collected in a participant's home during a brief visit by a research team member who could ensure adherence to a protocol that minimized sample contamination and dilution. Several thousand salivary samples were successfully collected across multiple sequential menstrual cycles from more than 200 rural Bolivian women.

C.4. Suitable Statistical Methods

The median or geometric mean of individual ratios is the better approach for computing mean UF (see Section 2.4.1 and Appendix B). The conclusions reported by Chatterton et al. (2006) were based on ratios of NW‐Bolivia and NW‐Chicago group means for P_Free‐SAL and P_Total‐VEN rather than on individual ratios. Some of the problems with this approach are evident in a scatterplot of their data (Figure 8D). The use of arithmetic means is inappropriate here, as “mean uptake fractions” computed in this way are generally not the same as the mean of the individual uptake fractions. (In contrast, if geometric means were used, the resulting uptake fractions would coincide with the geometric means of the individual uptake fractions.) For example, Table 4 compares P UFs calculated using Chatterton et al.'s (2006) method versus the arithmetic mean, geometric mean, and median of the individual UFs. Notice that the UFs computed with Chatterton et al.'s (2006) method are substantially different from those computed by the other methods. These differences are a reflection of the wide range of the hormone levels within each sample, and of the presence of outliers in the data.

As did Wang and Knyba (1985), it is very valuable to future investigators to present all data and measurements, even for those cases where the sample measurements were below assay sensitivity, and to calculate sample statistics that include and omit such measurements and other outliers.

Vitzthum, V. J. , Bellido D., Echalar L., Caceres E., and Thornburg J.. 2025. “Salivary/Serum Progesterone Ratio Differs Between Menstrual Cycle Phases but Not Between Populations: Implications for Health, Reproductive, and Behavioral Research.” American Journal of Human Biology 37, no. 7: e70077. 10.1002/ajhb.70077.

Data Availability Statement

The data that support the findings of this study are openly available in DRYAD at https://datadryad.org/.