Machine learning augments traditional fetal monitoring capabilities to enhance clinical efficiency, timely intervention, and standardization of care
By Emily Hamilton, MDCM
Electronic fetal monitoring (EFM) has been a mainstay in modern perinatal obstetrics since the 1970s for the detection of fetal distress during labor. The technology used in EFM has changed little since that time, although the demands on perinatal nurses—largely considered to be the leading providers and coordinators of labor and delivery (L&D) care—have escalated. Meanwhile, the use of technology in healthcare has seen exponential growth. Artificial intelligence (AI)—in particular, machine learning—is gaining prominence throughout medicine, and is now being applied to enhance fetal and maternal safety in L&D.
Today, there are many machine learning applications in medicine that perform pattern recognition, such as reading x-rays. In L&D, this technology is being applied to address the safety and well-being of babies and their expectant mothers. The use of machine learning provides consistent, objective, and data-driven analysis that can complement human abilities while compensating for lapses related to fatigue, stress, inexperience, and the well-documented nursing shortage.
It’s called labor for a reason
Labor introduces a variety of stresses on the unborn baby in the form of contractions and transitioning through a narrow passage. All babies experience a degree of oxygen deprivation (hypoxia) during contractions. It is usually mild and tolerated well. But if it is excessive or prolonged, or if the baby is not well equipped to deal with it, the resulting injury can be devastating—or deadly.
While the rates of hypoxia-related injury in the United States are among the lowest in the developed world, studies indicate that half or more of these outcomes are preventable and most of them are related to issues around fetal heart rate monitoring. This number is not unique to the United States; much of the Western world has completed studies on the subject that reached the same conclusion. Today, approximately one-third of women in the U.S. have their babies by cesarean section—a major surgery that carries risk. About half of cesareans in first-time mothers are related to fetal heart rate patterns, while the other half is mostly related to slow labor progress.
Most of those preventable hypoxic injury episodes relate to the human interpretation of fetal heart rate tracings, which raises the question: Why use these tracings to assess fetal oxygen status? Simply put, it is the only method we currently have access to. Maybe in the future there will be a better sensor or detector, but today, consistent interpretation of fetal heart rate tracings remains a difficult problem for clinicians who work under demanding conditions.
Although there are certain patterns in a fetal heart rate that indicate a baby is not tolerating labor well, it’s not a simple one-to-one relationship. There are a myriad of patterns, combinations, and trends over time that must be considered. How the clinician puts all of this information into context, given the other factors known about the patient, is key. It is an imperfect science at best.
Labor progression and maternal monitoring
All expectant mothers, and the clinicians taking care of them during labor, ask two fundamental questions: “How is the baby doing?” and “Is this baby coming out?” The analysis of fetal heart rate patterns is essential to assess how the baby is tolerating labor. But assessments of labor progress—determining whether the baby is descending, whether the uterus is opening as it should, and whether the baby will physically egress—are equally important.
During labor, nurses periodically measure the mother’s blood pressure, pulse, respiratory rate, and a few other indicators. Many of these measurements are monitored by machines, while others are manually measured and recorded. Assessing the mother’s status relative to the progress of labor is difficult because there are many factors to consider and no two mothers are identical. This is another problem that can be standardized using mathematical techniques.
Given the range of variables in monitoring the safety and well-being of babies and mothers, we increasingly rely on computers to analyze complex situations reliably, consistently, and quantitatively, rather than qualitatively.
Augmenting and supplementing clinicians
L&D nurses must examine and document a range of five to 10 characteristics in each tracing every 15–30 minutes, depending on the stage of labor. Granted, some of these measures are simple, such as counting the number of contractions a patient has had in the past 10 or 30 minutes. But there’s no reason a highly skilled nurse should be spending time counting contractions.
It’s well known that L&D clinicians work under difficult circumstances—often in the middle of the night, when they may have been awake for many hours and their judgment may not be at its best. Factor in multiple patients, the emotional overlay of treating multiple patients at once, and the clinician’s experience level. The nursing shortage spurred by the pandemic has not only hastened attrition but has necessitated extraordinary working hours. Nurses are exhausted and are being pulled in from other units to deal with the crisis and ensure coverage.
L&D nurses can readily analyze tracings—it’s the bread and butter of what they do every day. But it’s a challenge to do them consistently and efficiently and to show the trends, all while helping an inexperienced nurse or assisting a float nurse who’s been brought in from another unit. These situations pose a substantial risk for patients.
What if there were a reliable method that could analyze these tracings and other crucial measurements over time?
Recognizing patterns, consolidating information
Consider a computerized system that is not affected by fatigue, emotion, or patient load, and that can also standardize the interpretation of fetal heart rate monitoring. This is where applications powered by machine learning algorithms can provide an “extra set of eyes” in monitoring the unborn child’s condition and the mother’s labor progression.
Fetal heart rate monitoring is an ideal application for machine learning because it is a type of mathematics that involves ambiguity and differing presentations. Consider the variations in a handwritten signature—we never write our names exactly the same way, yet AI applications can recognize them. Fetal heart rate patterns are both variable and complex—not only because we examine what appears in a noisy background but because they evolve over time.
Today, machine learning is being applied to recognize, detect, label, and measure these fetal heart rate patterns. L&D systems powered by machine learning can consolidate up to 12 hours of monitoring information on a single screen—with color-coded status. Nurses and physicians can refer to this dashboard to quickly evaluate trends, including the state of the patient and the progression of labor.
Being able to see the trends in fetal heart rate patterns, maternal vital signs, and labor provides an idea of “if and when” a delivery is imminent, progressing normally, or in trouble. This approach also can provide the obstetrician with a more complete picture over time, as opposed to disjointed snapshots.
Apart from enabling standardization, precision, and consistency, machine learning applications in L&D substantially lower the workload for nurses, who not only must monitor and document their findings but also must consolidate and synthesize all that information into action.
For administrators, this capability also facilitates recruiting because new nurses will feel less isolated knowing that they have backup. As the pandemic continues, with some bedside nurses not able to work in L&D because of a personal health condition or risk of viral exposure, AI-powered monitoring systems enable the nurses to oversee the tracings from other locations. These nurses can support the bedside team and monitor tracings being recorded 20 feet or 20 miles away.
This new class of AI-powered technology works alongside existing perinatal systems, adding automated early warning capabilities to help clinicians identify concerning trends in their patients’ status. It is a good example of combining the best of two worlds. The technology side consolidates, analyzes, and trends data, highlighting deviation from expected norms. On the human side, clinicians apply their clinical knowledge and judgment to make well-informed decisions and take timely actions based on this clinical decision support.
Emily Hamilton, MDCM, is senior vice president of clinical research at PeriGen, a Halma company, which offers innovative perinatal software solutions that incorporate advanced statistical analysis features to enhance clinical efficiency and standardization of care during childbirth.