Era of Evidence-Based Medicine and the Artificial Intelligence: The Transition of Obstetric Decision-Making From Empirical to Precision-Based Approaches
Year: 2026
Type: article
Abstract: The traditional discipline of obstetrics is undergoing a transformation and is being revitalized. While classical theories, such as Friedman’s labor curve, are being challenged by new evidence, innovative technologies, including artificial intelligence (AI), multi-omics, and minimally invasive fetal surgery, are increasingly being integrated into clinical diagnosis and treatment strategies. With the advancement of new evidence-based medicine and novel technologies, obstetric decision-making is shifting from an empirical approach to a precision-based one. The delivery room, as the core clinical area in obstetrics, involves rapid and uncertain changes in the patient’s condition driven by regular uterine contractions. Physicians and midwives must remain vigilant at all times and actively employ innovative methods to make timely, comprehensive, and anticipatory decisions guided by evidence-based medicine and clinical guidelines. Only through individualized and precise management of each patient can optimal maternal and neonatal outcomes be achieved. Induction of labor strategies and tachysystole Obstetrics seeks to answer two classic questions: first, “When” to deliver, and second, “How” to deliver. Pregnant women with various comorbidities and complications require timely termination of pregnancy at appropriate gestational ages to reduce adverse maternal and neonatal outcomes. If labor does not commence spontaneously, induction of labor is necessary. In recent years, for low-risk nulliparous women, the ARRIVE trial in the United States1 found that induction at 39 weeks, compared with expectant management, did not significantly increase perinatal death or severe neonatal morbidity (relative risk (RR) = 0.80, 95% confidence interval (CI): 0.64–1.00) and was associated with a lower cesarean section rate (RR = 0.84, 95% CI: 0.76–0.93). While the generalizability of these findings remains controversial globally, the high-quality clinical research and systematic reviews emerging in the field of induction warrant attention. In implementing induction protocols, the chosen method should be tailored to the cervical condition. Misoprostol is a synthetic prostaglandin analog with stable effects and low cost. When administered vaginally, it effectively induces uterine contractions and improves cervical conditions, making it a commonly used induction agent. Cervical balloon dilation involves placing a balloon in the cervical canal to mechanically dilate it, thereby stimulating the release of endogenous prostaglandins, softening the cervix, and inducing contractions. Recent evidence-based medicine indicates that, when performed correctly, cervical balloon placement does not increase adverse pregnancy outcomes and effectively promotes cervical ripening. The combination of misoprostol and a cervical balloon is emerging as a potentially effective novel strategy for cervical ripening. A systematic review2 comparing induction with misoprostol alone versus misoprostol combined with a cervical balloon found that mechanical cervical dilation combined with misoprostol may accelerate the labor process and improve neonatal outcomes without increasing maternal or neonatal complications. Furthermore, labor management essentially involves the management of uterine contractions. Although the American College of Obstetricians and Gynecologists guideline defines tachysystole as > 5 contractions in 10 minutes, the need for intervention should be determined by integrating the patient’s contraction intensity and fetal heart rate patterns, balancing the need to progress labor with the avoidance of events such as fetal distress and precipitous delivery. Intrapartum ultrasound for predicting labor progress The use of intrapartum ultrasound has increased markedly in recent years, primarily for assessing labor progression. Intrapartum ultrasound is more accurate than traditional digital vaginal examination in evaluating fetal head position and descent. Additionally, ultrasound offers advantages such as being less invasive and better tolerated. Repeated ultrasound assessments can reduce the risk of intrauterine infection associated with frequent vaginal examinations. Ultrasound can also evaluate other labor parameters, including fetal head posture, the angle of head rotation within the birth canal, and asynclitism. Precise assessment of these parameters and implementation of individualized interventions can optimize maternal and neonatal outcomes and effectively reduce cesarean section rates. One study3 introduced the “occiput-spine angle” (OSA), an intrapartum ultrasound parameter for quantitatively calculating the degree of fetal head flexion relative to the spine during the first stage of labor. In 108 term singleton pregnancies, the mean OSA was 126° ± 9.8°. The risk of obstetric intervention (cesarean section or forceps/vacuum-assisted delivery) was significantly higher in women with OSA < 125° (odds ratio = 1.08, P = 0.04) and was significantly associated with prolonged labor (hazard ratio = 1.62, P = 0.02), suggesting that OSA is a reproducible and easily obtainable objective ultrasound parameter. A smaller OSA value (< 125°) indicates poor fetal head flexion and is an independent risk factor for predicting dystocia and operative delivery. Another study4 used ultrasound to measure fetal head position and flexion in women with arrested active-phase labor. For non-occiput posterior fetuses, the occiput-spine angle was measured; for occiput posterior fetuses, the chin-chest angle was measured. The results showed that fetal head deflexion was associated with an increased incidence of cesarean section due to labor arrest, suggesting that intrapartum ultrasound may help classify the etiology of labor arrest. A further RCT5 compared the use of ultrasound versus vaginal examination for assessing labor progress and found that intrapartum ultrasound reduced the total number of vaginal examinations required during labor, thereby lowering the incidence of intrapartum fever and chorioamnionitis. No adverse effects on labor progress or short-term maternal or neonatal outcomes were observed. AI, particularly deep learning, is rapidly expanding its application in obstetrics, especially in the field of intrapartum ultrasound. Deep learning technologies can significantly enhance the objectivity, reproducibility, and efficiency of fetal ultrasound, demonstrating superior performance in the automatic recognition of standard planes, discrimination of normal/abnormal anatomical structures, automated biometric measurements (e.g., biparietal diameter, head circumference, abdominal circumference, femur length, crown-rump length, nuchal translucency), and detection of anomalies in multiple fetal systems such as the central nervous system, heart, and placenta. Deep learning holds promise as a future decision-support tool to aid obstetricians in a standardized, real-time, and generalizable manner.6 Future AI models should aim to automatically analyze and integrate multiple ultrasound parameters to provide decision support for managing labor arrest, for instance, by assisting clinicians in making optimal choices between operative vaginal delivery and cesarean section. Interpretation of intrapartum fetal heart rate monitoring Following the release of the Chinese expert consensus on electronic fetal heart rate monitoring application in 2015, several international guidelines and consensus statements on fetal monitoring have been published, refining and updating many aspects. The 2022 Society of Obstetrician and Gynaecologists of Canada (SOGC), “Maternal Heart Rate Artefact During Intrapartum Fetal Health Surveillance”,7 defines maternal heart rate artefact (MHRA) as a condition in which the fetal monitor detects the maternal heart rate but erroneously outputs it as the fetal heart rate. A 2014 Canadian study8 showed that MHRA occurred at least once during labor in 721 (55%) cases, and in 35 (2.7%) cases, it could have led to misinterpretation; in 33 of these cases, MHRA masked an abnormal fetal heart rate (FHR), and in 2 cases, it masked a normal FHR. During labor, the maternal heart rate (MHR) baseline is typically lower than the FHR baseline. During uterine contractions or maternal pushing, the MHR may increase reflexively, increasing the proportion of MHR accelerations, which tend to have higher amplitude and longer duration than FHR accelerations. Conversely, during labor, FHR typically shows accelerations but a decreased frequency of decelerations. MHRA should be suspected when FHR changes or improves without an apparent reason, when accelerations occur concurrently with contractions and/or maternal pushing, or when there is an abrupt shift to a new baseline level or baseline variability. Furthermore, international expert consensus statement on physiological interpretation of cardiotocograph (CTG): first revision (2024)9 identified three typical cardiotocography (CTG) patterns requiring recognition and proactive intervention. The ZigZag pattern is defined as baseline variability > 25 bpm lasting > 1 minute. It results from autonomic instability (primarily parasympathetic) and is associated with an increased risk of neonatal acidosis and NICU admission (approximately 2-fold risk). Persistence beyond 1 minute requires interventions to improve fetal oxygenation (e.g., reducing oxytocin, inhibiting contractions, stopping maternal pushing). If baseline elevation occurs without decelerations but with a ZigZag pattern, clinicians should be vigilant for fetal neuroinflammation associated with chorioamnionitis. The suggestive of fetal inflammation (SOFI) pattern involves a baseline FHR elevation > 10% without repetitive decelerations, suggesting fetal inflammation and association with adverse perinatal outcomes. The relative utero-placental insufficiency of labour (RUPI-L) pattern is characterized by a rising FHR baseline during labor with widespread, deep decelerations coinciding with regular contractions; the width and depth of decelerations decrease or disappear when contraction intensity and frequency diminish. The pathophysiology involves impaired placental function leading to reduced placental oxygen reserve, with regular contractions further diminishing fetal oxygen supply. Due to chronic exposure to subclinical hypoxia, there is chronic release of adrenal-derived catecholamines, and the FHR baseline is typically at the upper limit of the normal range during intervals between decelerations. In summary, the interpretation of intrapartum fetal heart rate monitoring should be based on the type of fetal hypoxia and the fetal response to hypoxic stress, considering the underlying pathophysiology rather than mechanically applying pattern definitions. Clinicians should shift their focus from solely recognizing deceleration patterns to also evaluating baseline stability and variability. Furthermore, one study10 pointed out that noninvasive fetal electrocardiography can provide objective, beat-to-beat analysis of fetal heart rate variability, overcoming inherent limitations of traditional CTG, such as subjectivity, high body mass index dependency, and the inability to capture true beat-to-beat signals. It has demonstrated unique advantages in remote home monitoring, management of obese pregnant women, assessment of fetal growth restriction, and evaluation of cardiac arrhythmias. Although signal loss remains a major technical challenge—particularly between 26 and 34 weeks of gestation due to interference from vernix caseosa—the integration of AI algorithmsholds promise for improving the sensitivity of early hypoxia detection. From the individualized selection of induction strategies and the objective assessment of labor progress via intrapartum ultrasound, to the deeper interpretation of fetal heart rate monitoring moving from pattern recognition toward understanding pathophysiology, all these aspects reflect the ongoing transition in obstetric decision-making from empirical judgment to precision intervention, driven by both evidence-based medicine and AI. With the continuous maturation of multimodal data fusion and deep learning models, obstetrics is poised to advance further into a new phase of intelligent and individualized care. Funding Supported by Noncommunicable Chronic Diseases-National Science and Technology Major Project (2024ZD0532100) and National High Level Hospital Clinical Research Funding (Youth Clinical Research Project of Peking University First Hospital) (2025YC14). Conflicts of Interest None. Data Availability Data sharing not applicable to this article as no datasets were generated or analyzed during the current study. Editor Note Huixia Yang is one of the Editors-in-Chief of Maternal-Fetal Medicine. The article was subject to the journal’s standard procedures, with peer review handled independently of this editor and the associated research groups.
Source: Maternal-Fetal Medicine
Authors: Ye Feng, Huixia Yang
Institutions: Peking University, Peking University First Hospital
Language: English
Subfield: Obstetrics and Gynecology
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
Award: G3586891232
OpenAlex ID: W7152798166