The progression of human labor varies between individuals, which is associated with complicated physiological changes during pregnancy and labor (1). An important medical decision has to be made when a threatened labor (TL) patient visits the hospital, who might enter the latent phase within several hours, or wait for days before labor initiation. Currently, the evaluation for hospital admission is based on the clinical symptoms and cardiotocography (CTG) (2). However, the contractions and early symptoms might sometimes show inconsistency or only continue for a while, and the equipment might not be available in regions with limited medical resources. Thus, it would be helpful if additional test platforms and parameters could assist the decision-making in this emergency scenario.

In the past decades, efforts have been made to identify effective predictors for labor initiation (3–6). Stress regulatory hormones belong to the hypothalamic-pituitary-adrenal axis (HPA axis), corticotropin-releasing hormone (CRH), adreno-cortico-tropic-hormone (ACTH), and cortisol were all reported to be associated with parturition (7–10). Since the second trimester, CRH secreted from the placenta (pCRH) has increased dramatically (11). Maternal plasma CRH levels were significantly higher in preterm women who subsequently gave birth within 24 hours (12). The downstream ACTH and cortisol demonstrate a similar pattern and all three hormones peak during parturition, making them potential biomarkers for both preterm and term labor (13, 14).

As one of the most abundant hormones that directly regulate stress, metabolism, and fetal development, cortisol has been monitored for the prediction of labor in many clinical trials (15, 16). Prediction of preterm birth during 32-36 weeks’ gestation with plasma cortisol and other factors has been achieved with a decision tree model (17). Through a weekly assessment of saliva cortisol and other hormones, Alonso et al. (8) predicted the probability of spontaneous birth from 37-40 weeks gestation. The accuracy calculated from the random forest (RF) model of the full-term women ranges from 73.33% to 85.71%. Despite the improvement in prediction accuracy, both studies overlooked the gap between the latest time for cortisol evaluation (usually the last week before labor) and labor. It is unknown whether cortisol levels would increase during the expectant phase or after threatened labor with mild contractions and pain. Since the duration of the threatened labor phase varies from 1-2 hours to days, it could also be possible that serum cortisol is boosted within a short period before latent labor. Intriguingly, salivary cortisol concentration increased dramatically through different stages of labor, which was likely a reflection of the consistent physiological stress (18). Thus, we hypothesize that the rise of cortisol levels might be significant during the transfer between the TL and labor initiation stage, which is a vital period for doctors to make medical decisions.

To facilitate the diagnosis of labor initiation upon the arrival of TL patients, a fast and convenient onsite examination is preferred. However, hormone tests in previous clinical experiments were conducted afterward (18, 19). Blood CRH with low circulating concentration was mostly detected with radioimmunoassay, which involves radioactive elements and is time-consuming (17, 20). Measurement of ACTH and cortisol was often achieved with the enzyme or luminescence immunoassay, which requires hours for the detection or expensive equipment (8, 13). These limitations in detection tools also hinder the clinical application of the findings.

In this study, the dynamic change of serum cortisol during late pregnancy, TL, and labor initiation stages was investigated for full-term participants. Following the ordinary practice of the obstetric department, a novel point-of-care testing (POCT) platform was incorporated and enabled the rapid detection of cortisol onsite. To our knowledge, this is the first cross-sectional study that evaluates the dynamic change of serum cortisol between TL and labor initiation with a “come-and-test” study design. For clinical decision-making, the cutoff concentration of serum cortisol was calculated and multiple variables were applied to the machine learning algorithms for diagnosis. A clinical workflow was proposed based on current operations that help to make medical decisions.

With this onsite cross-sectional study, we found that the morning serum cortisol levels increased gradually during pregnancy. A sharp elevation of serum cortisol concentration was first reported within the relatively short period between TL and the latent phase, which could facilitate the diagnosis of labor initiation (AUC 0.934-0.940). The demographic information of patients might help to further improve the performance of the algorithms.

Different from previous longitudinal studies (8, 24, 25), interpersonal variation is included in this study to mimic the real situation of medical examinations, and the full-term patients were grouped based on physiological changes and symptoms (E, TL, and L) rather than a broad gestation age. Consistent with previous research, our study at the population level also revealed an increment of morning serum cortisol during T1 and T3 ( Table 1 ) (25). During late pregnancy (T3 to E), a graduate elevation of morning serum cortisol was observed (14.96-16.60 μg/dL), which was different from the significant increment of salivary cortisol reported by Alonso et al. after week 37 (8). This divergence may be due to the district secretion pattern of different tissues, or the relatively large gestation week variation in the T3 group. Based on these findings, we postulate that the elevation of cortisol secretion during pregnancy should be a common phenomenon in both preterm and full-term women (8, 24).

Interestingly, cortisol levels in the E group were found similar to those in the TL group, although the latter demonstrated typical symptoms and pain. To our knowledge, no previous study has compared the cortisol levels for these two groups. Between the E and TL groups, there was a clear difference in sampling time: blood was drawn at the “high point” of the day for the E group (26, 27), and randomly for the TL group. However, no correlation was identified between the blood sampling time and cortisol level, probably because the neuroendocrine change caused by the stress and pain of TL has already overcome the diurnal variation (24).

Within the short period between TL and L, serum cortisol levels nearly doubled (16.57-31.04 μg/dL), but no disparity in gestation week was found between group L, E, or TL. As verified through correlation tests, age, and BMI were not related to cortisol levels. Therefore, the differences in cortisol levels among the three groups were not likely caused by variations in gestation week, age, or BMI, but caused by the feedforward cycle generated by pCRH (along with ACTH and cortisol) which peaks during parturition (28). There could also be confounding factors that trigger the release of extra cortisol, such as physical or mental stress (pain, fear, or anxiety) and mild endocrine alterations (14, 18). However, our previous study with the same cortisol test identified only a slightly higher level of cortisol in women with mild to severe anxiety (29), suggesting that the major driving force of the extremely high cortisol level should be labor initiation. The labor group of previous studies included preterm patients of different conditions: some showed little symptoms (like our E group), but some with dilation of 2-4 cm (17, 30). In a labor group with different physiological conditions, the sharp increase of cortisol might be overlooked.

To verify the performance of serum cortisol level as a biomarker for labor initiation, ROC curves were constructed for groups TL and L, as well as E+TL and L ( Figure 2 ). The aim was to distinguish labor initiation within a short period (TL to L) or at a similar gestational week (E+TL and L). ROC based on serum cortisol levels demonstrated similar performance in both groups (AUC 0.934-0.940), and cutoff values (21.46 μg/dL and 21.62 μg/dL), suggesting that serum cortisol levels were effective for the diagnosis of labor.

Labor diagnosis with a binary logistic regression model incorporated cortisol and age from the background information ( Table 2 ). However, the age of patients in group TL was generally higher than that of group L. Therefore, we postulate that the inclusion of age might due to the convenience sampling bias. Besides, the correlation test demonstrated that age and cortisol levels were not related. The decision curves that provided a comprehensive representation of the strengths and weaknesses of classification methods implied a better risk and beneficial outcome for logistic regression analysis, possibly due to a better core algorithm ( Figure 3 ).

Diagnosis of birth with machine learning methods such as Decision Tree (DT) or RF has been verified in previous studies (8, 17), incorporating multiple key factors and improving the prediction power of small datasets. Variables such as maternal age, gestation week, and cortisol level were used in common with previous studies, while parity and BMI before pregnancy were adopted only in this study (31). As shown in Table 3 , the accuracy of the machine learning models was similar to that of the diagnosis based on the single indicator cortisol. With fewer variables used in the statistic model (cortisol only), the sensitivity and NPV were slightly higher (86.50% and 93.57%). In contrast, the 10-fold cross-validation of the SVM algorithm generated higher specificity and PPV (89.29% and 81.09%, Table 3 ). The similarity in confusion matrices and smooth calibration curves for the prediction algorithms suggested no overfitting for the machine learning models ( Figure 4 ). Since labor initiation generally involves complicated physiological changes that are affected by many factors, processing of high-dimensional independent variables is preferred. More data are needed to further improve the performance of diagnosis, especially for the application of machine learning methods. Additionally, evaluation of the proposed management algorithms with larger population sizes and more variables is also an important work in the future.

Despite that cortisol has been used for the prediction of parturition in several studies, few mentioned the rapid change of cortisol between TL and L or tried to evaluate labor initiation with POCT (8, 17, 24). This cross-sectional study is the first to track serum cortisol levels onsite through different stages of pregnancy and distinguish TL from labor initiation. The sharp increase in serum cortisol level was likely caused by stress, pain, and part of the complicated physiological and psychological change of labor (18, 32). Based on the result of our study and the research of Miller et al. on salivary cortisol concentration through the course of labor, we postulate that serum cortisol might further increase in later phases of labor and be maintained at a high concentration for a short period postpartum (18). The rapid increase of serum cortisol levels, together with the age, gestation week, parity, and BMI of the patients could help to diagnose labor initiation effectively.

Based on the practice of our hospital and this study, a workflow for TL/L diagnosis with the addition of serum cortisol tests was proposed ( Figure 5 ). Since the clinical practice between hospitals and cultures varies, this workflow aims to guide if an additional parameter, cortisol concentration, is needed. Patients with TL symptoms could go through 3 types of examinations upon arrival: basic information collection, CTG monitoring, and blood tests. These operations could be conducted in parallel for time-saving. If both the contraction and cortisol levels are low, the admission might not be necessary. A strong contraction and high cortisol concentration might indicate an emergency. Serum cortisol tests can also be conducted after regular clinical examinations if the condition of patients is inconsistent. In special situations where the patient proceeds to the labor phase very quickly, or CTG and professionals are not available (i.e., in an ambulance), a quick, convenient cortisol test might help to make a preliminary decision.

The special design of this study includes the application of a POCT platform for onsite cortisol assessment. The ordinary examination procedures of the hospital were followed and the proposed methods could be transferred to clinical applications. The rapid increase of stress hormones was found from TL to labor initiation even with interpersonal variations between groups. Additionally, our study compared the serum cortisol levels of E, TL, and L group patients of similar gestation age, which was overlooked in previous studies (8, 18). Taking the design and detection tool of this study, future works should reveal the dynamic change of cortisol at different stages of labor and thereafter (33). In particular, it might provide insight into the prolonged latent phase or arrested active phase issues (34, 35).

A few limitations of our study should also be discussed to better interpret the results. Firstly, the sample size was effective in demonstrating the difference between TL and L groups, but relatively small for the application of machine learning algorithms. Following the standardized protocol, the tests and analysis need to be validated in more clinical sites, including a larger population with enough diversity, and more variables as potential influencing factors (22, 36). The generalizability across populations and the feasibility of the proposed machine learning method and workflow should be further verified. Secondly, due to the examination schedule of local patients, data from the T2 gestation period were missing. Fortunately, this information could be found in the study of Braithwaite et al., where pregnant women in the 3rd trimester had higher cortisol levels than in the 2nd trimester in both normal controls and depressed populations (37). Thirdly, variations might exist within the labor group, therefore future studies should evaluate different stages of labor.

In conclusion, we found that the E and TL patients demonstrated a similar cortisol concentration despite the differences in clinical symptoms. A sharp increase in serum cortisol level occurred within the relatively short period between TL and L, indicating a final countdown towards parturition. The serum cortisol level and basic information of patients could help to diagnose labor initiation with high accuracy. Using serum cortisol concentration as an additional biomarker for labor initiation may help decision-making and improve the healthcare of patients.

In the future, other investigations that apply cortisol measurements in the case of preterm birth could be attempted. Although preterm birth was not within the scope of this study, the non-invasive cortisol test in this study was quite convenient, making it easy to evaluate preterm birth, especially for those who visit the hospital in an emergency. The method of POCT, experimental design, and statistical analysis can be applied, and more comprehensive background information such as blood pressure and glucose level can be integrated. The physiological mechanisms underlying the observed rapid cortisol change should also be explored, including the testing of pCRH and ACTH, two upstream regulators of cortisol, at the same time point. This will further reveal the timely regulatory mechanism of hormones during TL and labor initiation, providing valuable insight into the utility of cortisol in clinical practice.

This is the first cross-sectional study that focuses on the dynamic change of serum cortisol between TL and labor initiation, revealing the nearly doubled serum cortisol concentration between TL and L patients. A convenient POCT platform for serum cortisol was adopted, enabling the onsite rapid assessment of serum cortisol. With the same gestation week, the E and TL patients showed similar cortisol levels, although the physiological conditions were different. Conventional statistical analysis and machine learning algorithms were used for the distinguishment of threatened labor and labor with cortisol concentrations and multiple variables, reaching accuracy levels of 85.93%-89.79%. Thus, serum cortisol as an additional parameter might help to diagnose labor initiation and make medical decisions.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by the institutional review board of the Second Affiliated Hospital of Chongqing Medical University (clinical trial registration number: SAHCQMU 2021 No. 22.). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

YC: Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Resources, Supervision, Writing – original draft, Writing – review & editing. HW: Data curation, Project administration, Software, Writing – original draft. DT: Data curation, Investigation, Writing – original draft. YW: Data curation, Investigation, Writing – original draft. ZS: Formal analysis, Investigation, Writing – original draft. YZ: Investigation, Validation, Writing – original draft. BZ: Investigation, Validation, Writing – original draft. BN: Data curation, Funding acquisition, Project administration, Resources, Software, Writing – original draft. XD: Conceptualization, Funding acquisition, Investigation, Resources, Writing – original draft, Writing – review & editing.