Background
Biochemical cervical change during labor is not well understood, in part, because of a dearth of technologies capable of safely probing the pregnant cervix in vivo. The need for such a technology is 2-fold: (1) to gain a mechanistic understanding of the cervical ripening and dilation process and (2) to provide an objective method for evaluating the cervical state to guide clinical decision-making. Raman spectroscopy demonstrates the potential to meet this need, as it is a noninvasive optical technique that can sensitively detect alterations in tissue components, such as extracellular matrix proteins, lipids, nucleic acids, and blood, which have been previously established to change during the cervical remodeling process.
Objective
We sought to demonstrate that Raman spectroscopy can longitudinally monitor biochemical changes in the laboring cervix to identify spectral markers of impending parturition.
Study Design
Overall, 30 pregnant participants undergoing either spontaneous or induced labor were recruited. The Raman spectra were acquired in vivo at 4-hour intervals throughout labor until rupture of membranes using a Raman system with a fiber-optic probe. Linear mixed-effects models were used to determine significant ( P <.05) changes in peak intensities or peak ratios as a function of time to delivery in the study population. A nonnegative least-squares biochemical model was used to extract the changing contributions of specific molecule classes over time.
Results
We detected multiple biochemical changes during labor, including (1) significant decreases in Raman spectral features associated with collagen and other extracellular matrix proteins ( P =.0054) attributed to collagen dispersion, (2) an increase in spectral features associated with blood ( P =.0372), and (3) an increase in features indicative of lipid-based molecules ( P =.0273). The nonnegative least-squares model revealed a decrease in collagen contribution with time to delivery, an increase in blood contribution, and a change in lipid contribution.
Conclusion
Our findings have demonstrated that in vivo Raman spectroscopy is sensitive to multiple biochemical remodeling changes in the cervix during labor. Furthermore, in vivo Raman spectroscopy may be a valuable noninvasive tool for objectively evaluating the cervix to potentially guide clinical management of labor.
Introduction
Why was this study conducted?
This study was conducted to identify biochemical changes that occur in the human cervix during spontaneous and induced labor, characterizing the late ripening and dilation phases of the cervical remodeling process.
Key findings
This study found that the progression of labor is characterized by (1) increased collagen dispersion, (2) increased lipid-associated spectral signatures, and (3) increased blood-related spectral signatures. These changes correlate with time to delivery and cervical effacement.
What does this add to what is known?
This study has built on knowledge of the cervical remodeling process from animal models and ex vivo human studies using a safe and effective tool for monitoring in vivo cervical change during parturition. This study continued our previous work using Raman spectroscopy to characterize biochemical changes during the softening and ripening phases that lead to labor. We anticipate that this technology will be a valuable adjunct for clinical labor management and eventually preterm labor prediction.
Given this dearth of cervical evaluation technologies, there is a need for quantitative and objective monitoring of labor progression. In current obstetrical practice, the Bishop score is used to predict the outcome of labor induction. A bimanual cervical examination determines cervical dilation, effacement, consistency, position, and station, which are used to calculate the Bishop score. This metric can vary widely from 1 provider to another, resulting in low sensitivity and specificity. Even when combined with transvaginal ultrasound measurement of cervical length, the Bishop score provides at most a 70% accuracy for predicting vaginal delivery. Historically, the progression of cervical dilation over time during labor has been benchmarked against the sigmoid-shaped Friedman labor curve, used to define normal labor and identify abnormal labor, which may require an intervention. However, subsequent studies showed this method of tracking normal labor progression to be inaccurate, which may have contributed to a 100% increase in cesarean delivery rate during the past 30 years. As such, improved tools for monitoring cervical remodeling are needed in the context of labor management to inform decisions, such as when to administer induction or augmentation agents and when to identify failure to progress objectively. Such a technology has the potential to reduce the rate of unnecessary cesarean delivery by making the diagnosis of arrest of dilation more precise.
Multiple minimally invasive tools are under investigation for objective evaluation of the cervix during human pregnancy and labor, including techniques based on optical, , electrical, ultrasonic, and photoacoustic , phenomena. Light-based methods offer many advantages, including collecting and analyzing data in real time, minimally invasive measurement, and sensitivity to multiple biochemical components. Raman spectroscopy is an inelastic light-scattering technique that probes vibrational modes of molecules within a sample, providing a biomolecular “fingerprint” of tissue. This approach can probe longitudinal changes in cervical components, such as collagen, elastin, lipids, blood, and water, and is sensitive to biochemical changes in the human cervix throughout pregnancy and after delivery. Previous work using Raman spectroscopy to investigate the human cervix revealed decreased signals related to extracellular matrix (ECM) proteins and increased blood signals as a function of gestational age, in addition to spectral differences based on parity and body mass index (BMI). Furthermore, the instrumentation for Raman spectroscopy has already been adapted to integrate with the bimanual cervical examinations routinely performed during labor. The biochemical specificity, safety, and clinical translatability of this approach make it well suited for objectively assessing cervical change during labor. In contrast to ultrasonic and other optical methods that provide primarily structural information or monitor 1 or 2 biochemical constituents, Raman spectroscopy can characterize many biochemical changes associated with the onset and progression of labor, thus providing a holistic picture of the cervical remodeling process.
Improved understanding of the role of the cervix in term and preterm deliveries, and both successful and failed labor, is crucial to improving the management of laboring women. This study aimed to characterize the biochemical features of term labor onset and progression, such that future studies may identify deviations in these features in cases of preterm, protracted, or failed labor. Furthermore, this work aimed to determine the biochemical fingerprint of cervical remodeling during latent and active labor. To this end, in vivo Raman spectroscopy was used to perform longitudinal cervical measurements of laboring women. Raman spectral changes were analyzed as a function of time to delivery, demonstrating an increase in peaks linked to collagen disorganization and increased features associated with blood and lipid-based molecules.
Methods
Clinical protocol
Overall, 30 pregnant women admitted to Vanderbilt University Hospital’s Labor and Delivery Unit for induction of labor or spontaneous labor were recruited and enrolled in a retrospective observational pilot study under a protocol approved by Vanderbilt University Medical Center’s Institutional Review Board (approval number 172109) following the guidelines of the Declaration of Helsinki. The primary provider determined whether a patient was eligible to participate in the study based on their medical assessment and strict inclusion and exclusion criteria shown in the Supplemental Table . Informed written consent was obtained from each patient before performing any measurements. Patient demographics, medical history, and methods and timing of labor augmentation were recorded ( Table ). Several induction or augmentation agents were used in our patient population: (1) misoprostol (Cytotec; Pfizer, New York, NY), a synthetic prostaglandin (prostaglandin E1 [PGE1]) analog applied directly on the cervix to promote ripening ; (2) intravenous oxytocin (Pitocin; Par Pharmaceutical, Chestnut Ridge, NY) administered to strengthen contractions ; (3) placement of a Foley bulb to mechanically dilate the cervix ; and (4) amniotomy (artificial rupture of membranes), which promotes prostaglandin release and increases fetal head engagement to mechanically dilate the cervix.
| Variable | n or mean±SD |
|---|---|
| Total patients | 30 |
| Age (y) | 27.00±5.44 |
| Gestational age at delivery (wk) | 39.42±1.23 |
| Birthweight (kg) | 3.26±0.44 |
| BMI at delivery (kg/m 2 ) | 29.30±8.10 |
| ≤30 | 12 |
| >30 | 18 |
| Race | |
| White | 22 |
| African American | 6 |
| Hispanic | 2 |
| Parity | |
| Nulliparous | 11 |
| Multiparous | 19 |
| Labor | |
| Spontaneous | 4 |
| Induction | 26 |
| Outcome | |
| Vaginal delivery | 23 |
| Cesarean delivery | 7 |
| Induction or augmentation methods | |
| Pitocin | 26 |
| Misoprostol | 15 |
| Foley bulb | 10 |
| Amniotomy | 17 |
| Comorbidities | |
| Preeclampsia | 2 |
| Gestational diabetes mellitus | 4 |
| Gestational hypertension | 1 |
| Chronic hypertension | 4 |
The following measurement protocol was used for each patient: the medical provider examined the cervix and cleaned any blood or mucus. Next, the Raman probe was placed in gentle contact with the cervix, and spectra were measured from 4 sites on the ectocervix at approximately 12-, 3-, 6-, and 9-o’clock positions using an integration time of 3 seconds per measurement. Although the amount of pressure applied was not standardized, previous work has demonstrated that mild or moderate pressure does not have a significant impact on acquired spectra. The current practice at Vanderbilt is to perform cervical examinations approximately every 4 hours when the cervix is <6 cm dilated and every 2 hours after the cervix is >6 cm dilated. Raman measurements were taken at the same time that standard bimanual cervical examinations were performed, and clinical providers took care not to obstruct the light path with their gloves. Once the membranes ruptured, Raman measurements were no longer acquired to reduce the risk of infection. Mathematical modeling and thermal measurements have shown that <1% of the laser light from the optical probe reaches 1 cm into the cervix, and it has been safely used to measure the cervix of both nonpregnant and pregnant women. ,
Raman system setup
A portable fiber-optic probe-based Raman spectroscopy system ( Figure 1 ) was used to acquire Raman spectra from patients during labor. The system consists of a diode laser (785 nm) coupled to a fiber-optic probe; an imaging spectrograph coupled to an air-cooled back-illuminated, deep-depletion charge coupled device (CCD) camera to collect and analyze Raman signal; and a computer to control the CCD camera. The custom fiber-optic probe incorporates in-line filtering within the tip of the probe to minimize interference from signals generated by the components used in the probe. Spectra were calibrated using standard protocols and processed for fluorescence subtraction using modified polynomial fitting and noise smoothing with a Savitzky-Golay filter. Spectra with an intensity >1000 calibrated units at 1440 cm -1 were determined to be predominantly adipose and were discarded.
A visually guided Raman spectroscopy probe was developed and validated during the study ( Figure 1 ). Therefore, the study population consisted of patients who either underwent a speculum examination (n=17) or were measured using the visually guided probe (n=13). No difference in measured spectra was found between the 2 probe configurations, although the probe was coded and incorporated as a fixed effect in the subsequent linear mixed-effects models to capture any potential variability introduced by the probe.
Data processing and analysis
Spectral measurements acquired from different locations on the cervix within the same patient at a given time point of labor were averaged, and the resulting spectra were plotted. Patients who ultimately delivered via cesarean delivery were excluded from subsequent analysis as a function of time to delivery, as their delivery timing is not dependent on cervical status, which this study sought to capture. The Raman spectra from the remaining patients were spectrally modeled as a function of time until delivery, starting 24 hours before delivery, in 4-hour increments. The overall longitudinal progression of cervical change was visualized using a restricted cubic spline with 65 degrees of freedom. Time of delivery was obtained from electronic medical record documentation.
Spectral features, which contributed significantly to the variation in the data as a function of time to delivery, were identified using the machine learning technique least absolute shrinkage and selection operator (LASSO). , To identify spectral changes throughout labor, linear mixed-effects models were used. This method is a type of general linear model that accounts for the correlation between repeated measures from the same patient, enabling identification of spectral features that change significantly ( P <.05) with time to delivery. General linear models follow the simple linear equation: Y = βX + ε, where Y is a vector containing the dependent variable (in this case, Raman spectral features), X is a matrix containing the independent variables (time to delivery, effacement, dilation, etc.), β is a vector containing coefficients of the independent variables (weights), and ε is the residual error in the model. A linear least-squares regression method was applied to determine β coefficients such that ε was minimized.
Major biochemical contributors to the Raman spectra at each time point were determined using a nonnegative least-squares (NNLS) biochemical model as previously described. Linear mixed-effects models were used to analyze how the spectral contribution of each molecule changed as labor progressed. All statistical analyses were performed using R software (R Foundation for Statistical Computing, Vienna, Austria).
Results
Patient population
A total of 30 participants were enrolled, and the demographics of this population are shown in the Table , with the median gestational age at delivery being 39.4 weeks. Of these patients, 26 were scheduled for induction of labor, whereas the remainder presented with spontaneous labor. Furthermore, 7 patients underwent unplanned cesarean delivery and were subsequently excluded from future analyses as a function of time to delivery, whereas the remaining patients had uncomplicated spontaneous vaginal deliveries. The study population included 11 nulliparous and 19 multiparous patients. Several methods of induction or augmentation of labor were implemented at the discretion of the clinical provider, and the number of patients who received each method is listed in the Table .
Spectral changes as delivery approaches
The time points at which measurements were acquired varied depending on the length of labor and timing of membrane rupture or other contraindications to study measurements. A spectral model was generated to approximate the expected spectra at evenly spaced time points. The spectral model shown in Figure 2 summarized the overall changes in the Raman spectra as patients approach delivery. Regions of interest in which observable changes in peak shape or intensity were noted were highlighted by shaded rectangles. Based on visual inspection of the spectral model alongside LASSO for feature selection, we observed decreased features associated with collagen and other ECM proteins (1248 cm -1 Amide III v (C-N), 1450 cm -1 CH 2 bending, and 1657 and 1668 cm -1 Amide I C=O stretch ). In addition, we saw an increase in signatures associated with lipids (1300 cm -1 CH 2 /CH 3 twisting and 1440 cm -1 CH 2 bending ), a decrease in spectral peaks associated with DNA (1339 cm -1 adenine and 1664 cm -1 C=O stretching modes of pyrimidine bases ) and actin (1340 cm -1 tryptophan, 1450 cm -1 CH 2 bending, and 1660 cm -1 Amide I C=O stretch ), and increased features associated with blood (1556 cm -1 C β C β vibration and 1620 cm -1 C a =C b vibration ). Furthermore, separate spectral models were generated for patients who received misoprostol for labor augmentation and those who did not receive the drug, shown in Supplemental Figure 1 . Key differences between patients who did not receive misoprostol and those who did included a more prominent 1339 cm -1 peak (DNA, P =.552). Moreover, a more dramatic decrease in the 1440 cm -1 and 1660 cm- 1 (ECM protein, P =.000005 and P =.098, respectively) peaks and a more pronounced increase in the 1556 cm -1 (blood, P =.59) peak were observed in the population without misoprostol. Although trends were observed related to misoprostol administration, most did not reach the level of statistical significance.
