Jan. 13, 2023
The ability for obstetricians and other medical professionals to quantify a patient's progression through labor as favorable or unfavorable is crucial for assurance that the process is proceeding well and to recognize quickly when intervention is necessary.
Abimbola O. Famuyide, M.B.B.S., is a gynecologic surgeon at Mayo Clinic's campus in Minnesota. Dr. Famuyide and colleagues found that a labor risk score (LRS) derived from machine learning can successfully provide a customized risk for a patient while it accounts for the evolving nature of labor. They published their findings in a 2022 issue of PLOS One. The study found this real-time method to be a potential viable alternative to traditional labor charts and models proposed over the years and useful for clinicians to determine what course to take in a delivery.
"What we hope is that obstetricians can utilize this tool to aid in balancing the neonatal and maternal risks of deferred intervention with the risk of unnecessary cesarean delivery," says Dr. Famuyide.
A study published in Vital Statistics Rapid Release in 2017 pointed to the fact that while cesarean delivery (CD) rates have risen dramatically in the last 30 years, maternal and neonatal adverse outcomes have not decreased correspondingly.
A 2002 study published in American Journal of Obstetrics and Gynecology by Jun Zhang, Ph.D., M.D., and colleagues proposed that the Friedman labor curve used since 1955 may no longer reflect patients' current labor patterns. Dr. Zhang and colleagues then conducted a prospective, multicenter study utilizing data from the Consortium on Safe Labor (CSL) database to test this hypothesis. They published findings in a 2010 issue of Obstetrics & Gynecology that confirmed contemporary labor patterns differed from the 1950s patterns. This study led to a new labor curve that altered obstetrical management in the U.S. and replaced the Friedman curve that had been used since 1955. However, Dr. Zhang's study excluded outcomes for mothers and neonates, which made Dr. Famuyide and colleagues question the utility of this labor curve for improving these outcomes and CD reduction.
"We wanted to build on the previous work of Dr. Zhang and colleagues and develop a labor chart individualized to the patient with prediction models that use machine-learning algorithms utilizing data on cesarean delivery and childbirth outcomes," says Dr. Famuyide.
How the study was conducted
The study by Dr. Famuyide and colleagues utilized anonymized existing data from the CSL database, which included 228,438 deliveries. The team developed intrapartum models for the prediction of unfavorable labor outcomes, which they defined as:
- Apgar score below 7 at five minutes postdelivery.
- Neonatal hypoxemic ischemic encephalopathy.
- Suspected or confirmed intra-amniotic infection.
- Admission to the neonatal intensive care unit.
- Neonatal death.
- Neonatal intracranial hemorrhage.
- Neonatal sepsis.
- Postpartum hemorrhage with estimated blood loss greater than 1,000 mL or need for blood product transfusion.
- Shoulder dystocia.
- Below 7.00 umbilical arterial pH.
- Unsuccessful vaginal delivery or CD in active labor.
- Need for neonatal ventilation or continuous positive airway pressure therapy.
The model determined an LRS for the probability of one of these outcomes. Dr. Famuyide and colleagues excluded women's cases with the following criteria:
- Active herpetic lesion.
- Pre-active labor CD, or CD performed at cervical dilation of at 5 cm or less.
- Cord prolapse.
- Elective CD.
- Inadequate documentation, or fewer than two cervical exams.
- CD for failed induction.
- Fetal anomalies.
- Fetal malpresentation.
- Intrauterine fetal death.
- Multifetal pregnancy.
- Preterm labor, or birth at less than 37 weeks' gestation.
- Three or more prior CDs.
Investigators created outcomes prediction models, including a baseline model and various intrapartum prediction models, which can accommodate a dynamic labor starting at 4 cm dilation. Researchers folded any variables unrelated to a certain dilation measurement into the prediction model for 10 cm dilation. They did not include fetal heart rate monitoring due to insufficient CSL database documentation. Statistical analyses utilized R programming environment for statistical computing version 3.5.1.
Of the 66,586 deliveries fitting eligibility criteria, 14,439 (21.68%) of the CSL database incurred unfavorable labor outcomes. For those who encountered unfavorable outcomes, baseline LRS measured more than 35%, while women who had favorable outcomes had a baseline LRS of less than 25%.
Utility of the LRS machine-learning model
Dr. Famuyide believes the machine-learning model currently undergoing additional validation studies may more effectively account for the nonstatic nature of labor, which he says was missing in the previous publications. It serves as an intrapartum prediction of risk that incorporates adverse birth and maternal and neonatal outcomes.
The new model gives the obstetrician the ability to evaluate the patient's baseline LRS, LRS trend over time and LRS graph relative to a reference LRS graph. Thus, the obstetrician can accurately predict unfavorable labor outcomes.
"The assumptions used in traditional statistical approaches are simply not universally applicable to both the changeability and the complexity of labor progression, while this machine-learning model can account for these factors," says Dr. Famuyide. "The predictive ability of a machine-learning model such as the LRS is strong. This model can improve the intrapartum decision-making process significantly."
For more information
Famuyide AO, et al. Impact of labor characteristics on maternal and neonatal outcomes of labor: A machine-learning model. PLOS ONE.17:e0273178.
Hamilton BE, et al. Births: Provisional data for 2016. Vital Statistics Rapid Release. 2017;2.
Zhang J, et al. Reassessing the labor curve in nulliparous women. American Journal of Obstetrics and Gynecology. 2002;187:824.
Zhang J, et al. Contemporary patterns of spontaneous labor with normal neonatal outcomes. Obstetrics & Gynecology. 2010; 116:1281.
Refer a patient to Mayo Clinic.