(C) PLOS One [1]. This unaltered content originally appeared in journals.plosone.org.
Licensed under Creative Commons Attribution (CC BY) license.
url:https://journals.plos.org/plosone/s/licenses-and-copyright

------------



Feasibility of continuous distal body temperature for passive, early pregnancy detection

['Azure Grant', 'The Helen Wills Neuroscience Institute', 'University Of California', 'Berkeley', 'California', 'United States Of America', 'Benjamin Smarr', 'Department Of Bioengineering', 'San Diego', 'Halicioğlu Institute For Data Science']

Date: 2022-05

Most American women become aware of pregnancy ~3–7 weeks after conceptive sex, and all must seek testing to confirm their pregnant status. The delay between conceptive sex and pregnancy awareness is often a time in which contraindicated behaviors take place. However, there is long standing evidence that passive, early pregnancy detection may be possible using body temperature. To address this possibility, we analyzed 30 individuals’ continuous distal body temperature (DBT) in the 180 days surrounding self-reported conceptive sex in comparison to self-reported pregnancy confirmation. Features of DBT nightly maxima changed rapidly following conceptive sex, reaching uniquely elevated values after a median of 5.5 ± 3.5 days, whereas individuals reported a positive pregnancy test result at a median of 14.5 ± 4.2 days. Together, we were able to generate a retrospective, hypothetical alert a median of 9 ± 3.9 days prior to the date at which individuals received a positive pregnancy test. Continuous temperature-derived features can provide early, passive indication of pregnancy onset. We propose these features for testing and refinement in clinical settings, and for exploration in large, diverse cohorts. The development of pregnancy detection using DBT may reduce the delay from conception to awareness and increase the agency of pregnant individuals.

Pregnancy impacts everyone, and is of major psychological importance to most. Early detection of pregnancy onset could be a major boon to women seeking to plan for the future and adjust their behavior (e.g. cessation of alcohol or drug use). At present, most women become aware of their pregnancy well after implantation, and support is often limited to periodic check-ups. Given the emergence of tools that allow for continuous monitoring of physiology (i.e. wearable devices), we feel that utilizing such tools could lead to rapid progress in developing safe, personalized insights across a woman’s fertility journey. Our work here does not aim to solve this problem, but to inspire others to appreciate the reality of this opportunity, and to consider using these tools to expand our understanding and assessment of pregnancy and related events.

Competing interests: The authors have read the journal’s policy and the authors of this manuscript have the following competing interests: AG and BS have both received compensation from Oura within the past year; AG in an internship, and BS as a scientific advisor. Oura did not fund this study, did not have the opportunity to review the data collected during this study, and has not had the opportunity to alter this manuscript in any way.

Data Availability: Data Availability and Cleaning. Code for statistics and feature generation are available at A.G.'s github. Raw data are proprietary under the Data Use Agreement between UCSD and OuraRing, Inc. Access can be requested of the corresponding author, but must be reviewed on a case-by-case basis by OuraRing, Inc. Contact: [email protected] .

For example, murine elevation of continuous body temperature can be used to detect pregnancy onset within ~12 h of conception [ 41 ]. Subsequent investigations in humans revealed continuous temperature-based predictors of the preovulatory LH surge [ 42 ], as well as prediction and detection of sickness [ 88 ]. Based on this growing body of work and on community reports, we hypothesized that features extracted from continuous distal body temperature at the finger (DBT) could provide early indicators of pregnancy. Here we report results from retrospective wearable temperature data from 3 months prior to 3 months after self-reported conception and compared this to self-reported dates of OTC or clinically confirmed pregnancy.

Elevation of basal body temperature (BBT) during pregnancy, and its putative relationship to progesterone, has been recognized for about a century [ 47 – 68 ]. However, BBT has proven too imprecise and difficult to collect for development as a reliable method of either pre-ovulatory luteinizing hormone surge [ 69 – 71 ], ovulation [ 70 – 74 ], or early pregnancy detection [ 75 – 78 ]. Despite this, community groups [ 79 ] frequently adapt BBT-based methods of ovulatory cycle tracking (i.e., the sympto-thermal method) in an attempt to detect pregnancy onset [ 80 , 81 ]. Although efforts have been made to automate the collection of BBT or similar metrics, they have focused largely on proception or contraception rather than on pregnancy detection [ 82 – 85 ]. Continuous (as opposed to once per day) collection of temperature is a promising alternative that appears to provide additional information about reproductive status by examining not only levels but carefully selected patterns of temperature [ 40 – 42 , 86 , 87 ].

There is evidence to suggest that passively measurable outputs with direct mechanistic ties to female reproductive physiology, such as continuous body temperature, could advance pregnancy testing [ 40 , 41 ]. A growing number of wearable sensors offer non-invasive, continuous body temperature measures [ 40 , 42 – 44 ]. Such timeseries, if appropriately measured and analyzed [ 42 , 45 , 46 ], provide a window onto reproductive events with great temporal granularity, and create an opportunity for precise mapping of patterns of change in pregnancy [ 40 ].

The development of automated tools has been limited by our low temporal resolution understanding of somatic changes in early pregnancy, and an historical reliance on single time-point hormone samples. Current clinical or over-the-counter (OTC) pregnancy tests rely on serum or urine measurements of human chorionic gonadotropin (hCG), and claim to detect elevated hCG with 99% accuracy at approximately the date of missed menses [ 34 ]. However, independent studies place most at-home test kit accuracies far lower, even weeks after missed menses [ 34 – 37 ]. Additionally, as pregnancy test efficacy is measured in days relative to expected missed period (i.e., about 12–17 days after conception [ 38 ]), the number of days post-conception that a test becomes accurate is both difficult to estimate [ 38 ] and likely more variable for the ~1/3 of women with irregular or long menstrual cycles [ 39 ]. Although OTC testing provides a potentially early indicator, most individuals who seek testing do not do so until weeks after a positive test may be obtained [ 1 – 3 ]. For unplanned pregnancies, this delay may be even more prolonged.

Automated, early pregnancy detection may substantially improve maternal and fetal health. Although population reports of pregnancy testing behavior are sparse [ 1 – 3 ], available cohorts indicate that most American women confirm pregnancy about 3 to 5 weeks after conception, and that nearly a quarter do so later than 5 weeks [ 1 ]. The cost of delayed pregnancy awareness is great: adoption of appropriate pregnancy habits is delayed [ 4 , 5 ] in the earliest, most fragile developmental window [ 6 – 9 ]. Moreover, unplanned pregnancy occurs at a rate of ~50% [ 10 – 13 ], and is associated with higher likelihood of adverse exposures [ 4 , 5 , 14 , 15 ], increased maternal morbidity and mortality [ 16 , 17 ], preterm birth and low childhood weight [ 18 – 21 ], elevated risk of birth defects [ 22 , 23 ], and poorer maternal psychological health [ 24 , 25 ]. As the effective window for emergency contraception is ~120 h at most [ 26 – 29 ], and as access to pregnancy testing and safe abortion access continues to be limited around the world [ 16 , 30 ], delayed confirmation poses considerable risks to pregnant individuals [ 16 , 17 ]. Early, passive pregnancy detection could increase the agency of a pregnant individual, speed adoption of pregnancy-safe behaviors (e.g., avoidance of environmental risk factors [ 7 ], cessation of alcohol consumption [ 8 , 31 ] or drug use [ 32 ]), or provide the choice to discontinue a pregnancy at an earlier gestational age [ 33 ].

We previously published the use of features derived from continuous core body temperature to detect pregnancies in mice, and to monitor pregnancies through delivery [ 41 ]. Here we highlighted a comparison of a mouse and a human mother across gestation ( Fig 4 ) to support future investigations of continuous body temperature patterns beyond conception. Despite obvious differences in gestational length, body size, and the species’ reproductive physiology, we noted several striking visual similarities between the two pregnancies. Both exhibited ovulatory cycles, followed by a steep rise in temperature at conception, and a long slow decrease in temperature until around the 3 rd trimester. Then, large elevations appeared in the temperature trajectory, followed by another steep rise preceding delivery. The consistency and mechanisms of these changes are beyond the scope of this manuscript. We hope that sharing this visual inspection is sufficient support the apparent existence of these conserved patterns in DBT and support more research in the field.

In order to explore the feasibility of enhancing contrast at transitions from follicular to luteal phases, and from follicular phases to pregnancies, we generated second order features normalized to the average historical length of the follicular phase (17 days [ 39 ]) (see: Methods for discussion of normalization window). Normalization of an individual’s nightly maxima to the previous 17 days’ history of minima ( Fig 3B and 3C red), and medians ( Fig 3B and 3C dark blue) exhibited faster within-individual statistical elevations above luteal highs than did Nightly maxima. The derived feature comparing each night’s maximum to historical minima and medians reached statistically significant elevation after 6 days (relative to historical minima: χ 2 = 78.00, p = 3.59*10 −12 , L vs. C +6 days p = 0.034, L +7 days and onward p< 1.60*10 −3 ; relative to historical medians: χ 2 = 91.0, p = 1.08*10 −14 , L vs. C +6 days p = 0.040, L +7 days and onward p< 4*10 −3 ). By contrast, while comparison to historical nightly maxima ( Fig 3B and 3C gold) had a significant effect of time, no individual day achieved significance after reported conceptive sex within the days assessed (χ 2 = 56.74, p = 3.72*10 −8 , L vs. C p>0.05 on all days). Although multiple features reached group significance, on the same day, different individuals reached each feature at different times, as shown in the histogram of RHAs ( Fig 3D ).

(A) Median nightly temperature maxima ± MAD (see: Methods ). (B) Overlay of the three features derived from comparison to individuals’ historical medians (blue), minima (magenta), and maxima (gold) (see: Methods ). Post-hoc tests of the difference between putative luteal-phase peaks, and the difference between putative-luteal phase peaks (“L”) and days from conception (“C”), indicated significant post-conception elevation on day +8 in nightly maximum, day +6 in derived minimum, day +6 in derived median, and on no days in derived maximum. Linear (D, top) and cumulative (D, bottom) histograms of days on which these features generated an RHA (see: Methods ).

(A) Median raw minutely temperature records from all 30 individuals, from 2 months before to 2 months after reported conception. (B) Histogram of pre-conception temperatures in (A) revealed a bimodal distribution, reflecting day-night oscillations, as seen more clearly in a zoom of one arbitrary week (C). (D) Data from A, zoomed to highlight the nightly highs (top) and daily lows (bottom). (E) Population median of the nightly temperature maxima (see: Methods ) enhanced the clarity of the patterns reflecting putative ovulatory cycles and pregnancy onset contained within DBT.

Individuals’ nightly maximum temperatures, aligned to the date of reported conception (grey) for all 30 participants. Red: Time between reported conception and reported test confirmation is shown in red unless (blue) an RHA occurred for that individual, at which point red is replaced by blue. At the date of test confirmation, records return to grey. Bottom: red vectors represent mean time from conception to confirmation for: the U.S. population (3.5 weeks); vulnerable sub populations of “Black and Hispanic” (3.9 weeks) and “Teen” (4.6 weeks), and the 23% of the U.S. population reporting “late confirmation” of more than 5 weeks after conception. Epidemiological data visualized from (1).

Features of DBT exhibited unique elevations following self-reported date of conceptive sex (see: Methods ) in all 30 cases, as well as reproducing previous reports of approximately monthly patterning associated with the ovulatory cycle [ 42 , 43 ] ( Fig 1 ). Unique DBT elevations were used to create a Retrospective Hypothetical Alert (RHA) (see: Methods ) that was triggered before standard positive pregnancy test results in 29/30 cases ( Fig 1 ). This is illustrated in Fig 1 as the transition from unconfirmed pregnancy following conception (red) into RHA-tagged pregnancy (blue), which precedes the transition back to grey that occurs upon confirmatory testing. RHA occurred a median duration of 5.5 ± 3.5 days after reported conception, whereas reported confirmation using a standard pregnancy test occurred after a median of 14.5 ± 4.2 days. The RHA-confirmed days occurred significantly earlier than pregnancy test confirmation (p = 1.05*10 −8 ). To provide context for the potential utility of such passively-generated alerts, we highlighted comparison to average pregnancy confirmation dates, including the U.S. population at large, as well as two populations that are at higher risk of pregnancy complications, “Black and Hispanic” [ 89 , 90 ] and “Teen” [ 1 , 16 ]. Additionally, we highlighted the estimated 23% of U.S. pregnancies which are not confirmed by approximately 5 weeks after conception [ 1 ].

The thirty individuals in this feasibility cohort ranged from 25–44 years old. Ten percent were 25–29, 50% were 30–34, 23% were 35–39, and 17% were 40–44. Of all individuals, 80% discovered their pregnancy through an over the counter (OTC) test, 7% through an in-clinic test, and 13% reported an IVF-initiated pregnancy. Pregnancies in this cohort were confirmed between May of 2019 and April 2021.

Discussion

This preliminary study supports the feasibility of passive, early pregnancy detection using continuous DBT. In this cohort, nightly temperature maxima rose rapidly in early pregnancy and reached uniquely high values an average of 5.5 days after self-reported conceptive sex. These features enabled generation of RHAs for all individuals, which illustrated that such approaches have the potential to cut time to pregnancy awareness, in this case by 2/3, as compared to reported use of pregnancy tests [1]. Additional analysis of large, diverse cohorts is necessary to understand population variability in DBT patterns and to appropriately refine the features described here before attempting to create prospective alerts. Furthermore, preliminary comparison to published murine data suggests that additional features of DBT may be useful for pregnancy applications beyond conception.

The potential utility of temperature measurement for pregnancy detection was published nearly 100 years ago [47,48,92]. Van de Velde reported a sharp rise in daily oral basal body temperature (BBT) in early pregnancy [47,48], and Buxton and Atkinson subsequently postulated its dependence on elevation of progesterone [49]. Amazingly, even finger temperature was proposed in 1949 by Burt, as a method of pregnancy monitoring [93] (although the authors were not able to locate primary follow-up studies [94–98]). Despite this, neither oral thermometry nor peripheral temperature have developed into trusted tools for early pregnancy detection [75–78,99].

Why might continuous distal temperature improve upon historical uses of oral temperature? Although skin temperature measurements taken from poorly vascularized areas may reflect environmental influences, DBT taken from the ventral portion of the fingers reflects autonomic influence on the brachial artery [100,101], and its patterns of change are not predominantly determined by an individual’s temperature environment [91]. Temperature at the periphery of the body also exhibits higher amplitude circadian and ultradian rhythms than temperature taken from the core (e.g., > 10°C vs. ~ 2–3°C), meaning that small variability in sensor accuracy or precision is less likely to perturb patterns of change assessed at the periphery [102,103]. Additionally, the greater comfort of continuous wear of a ring, as opposed to an alternative to capture core temperature, may enable the collection of data with fewer gaps, which is essential for accurately assessing temperature shifts occurring at non-standard time of day within and across individuals. As continuous measurement allows more information to be captured per unit time, DBT provides a much wider range of potential features associated with adaptation to a pregnancy than does once-per-day BBT [40,104]. For example, continuous DBT features have recently been used to anticipate LH surge onset and assist in cycle phase classification [42,43], and well as contribute to the inference of sleep-wake status and sleep stage [45,91,105,106]. Finally, DBT is temporally coupled to physiological systems that rise in output across the first trimester of pregnancy, such as progesterone [107,108], night time heart rate [109], insulin activity [110–112], luteinizing hormone [42], and cortisol [113,114].

This additional temporal structure, and more thoroughly mapped relation to hormonal output, allows intentional selection of timeseries features that reflect physiological changes. In this case, sleeping temperature measurement provides a relatively unperturbed state (akin to the state that first morning BBT roughly attempts to capture), in which an individual’s behaviors (e.g., exercise, outdoor environment) may have a more limited effect on temperature level and pattern. It is possible that this relative absence of behavioral contributions enables endocrine contributions to the DBT signal to take precedence [115,116]. As pregnancy is associated with substantial changes to hormone levels and thermoregulation, as well as cardiovascular remodeling, future research is needed to confirm that relationships between temperature and hormonal status hold in pregnant individuals.

Notably, the authors were unable to find high temporal resolution descriptions of peripheral physiology around the time of conception. It is possible that the present lack of tools for conception detection makes the study of very early natural pregnancies challenging. Some of the features we explored were more effective than others, and larger cohorts will likely reveal that some features work better in some groups than in others. If and when these or related features are developed to the point that they reliably provide information in larger cohorts, then continuous DBT may enable more precise study of the physiological changes associated with conception, including continuous glucose patterning [110,111], salivary and urinary hormone output [117,118], early cardiovascular remodeling [119]. Additionally, continuous data may allow definition of tighter critical windows for the impacts from maternal and environmental influences [8]. Much remains to be learned about physiological diversity in continuous DBT, and there are likely myriad strategies for generating predictive features. The present study supports the hypothesis that wearable devices can be used to map this physiological diversity and identify features that provide useful information on an individualized basis.

Such tools are not without risk and require careful validation at scale. False positives could result in unnecessary psychological stress or unwarranted consumption of emergency contraception [120], whereas false negatives could lead to a lack of caution in a true early gestation. Additionally, there are several limitations to a retrospective study based on wearable device data and fallible self-report [121,122]. For instance, self-report of day of conception is prone to error, as many individuals have sex multiple times during the fertile window preceding pregnancy onset [74]. Additionally, the retrospective design of the present study meant that exact dates of follicular and luteal phases prior to pregnancy onset could not be determined, necessitating the use of previously published DBT features for estimating luteal phase temperature peaks [42,43,72]. Future studies that include additional non-pregnant cycles from a larger and more diverse population are needed to obtain reliable false positive and false negative rates of detection based on the features described here, as well as the duration after conceptive sex necessary to differentiate pregnancy from other potential sources of abnormal temperature elevation. Indeed, it is possible that transient sickness, medication, metabolic or weight changes could contribute to individuals’ nighttime temperature; however, such modulations would be likely to disrupt, rather than reinforce, temperature patterns associated with ovulatory cycles and pregnancy. This study did not differentiate pregnancies that go on to have second trimester miscarriage, stillbirth, or other complications, and future work is needed to uncover if unique conception profiles are associated with these outcomes [47,78]. Moreover, health algorithms have a history of bias and failure to generalize [123], and our cohort likely represents a relatively affluent and health literate sector of society, which cannot be taken as broadly representative. For these reasons, our work supports the need for large, clinical studies on diverse populations to determine which parameters may provide precise indication of conceptive sex, fertilization, and subsequent events. More broadly, our work supports exploration of physiological diversity from timeseries data.

Our analyses of continuous DBT demonstrate the feasibility of passive, early pregnancy screening. Confirmation and augmentation of these features in clinical settings and across large, heterogeneous populations may enable development of wearable tools for pregnancy monitoring. Such tools may increase individuals’ agency and guide pregnancy testing at an earlier gestational age [8].

[END]

[1] Url: https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000034

(C) Plos One. "Accelerating the publication of peer-reviewed science."
Licensed under Creative Commons Attribution (CC BY 4.0)
URL: https://creativecommons.org/licenses/by/4.0/


via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.