Electronic fetal monitoring (EFM) has aided intrapartum fetal surveillance for more than four decades. In spite of numerous trials comparing EFM with standard fetal heart rate (FHR) auscultation, it remains unclear that this modality has led to improved perinatal outcomes, especially lower rates of perinatal morbidity and mortality. A variety of ancillary methods have been developed to improve the accuracy of EFM for predicting fetal compromise. At present, a limited number of studies have shown that the addition of fetal electrocardiogram (ECG) analysis to visual interpretation of FHR patterns resulted in better fetal outcomes. However, the shortcomings of visual interpretation of FHR patterns persist. Although automated systems for FHR analysis have been developed, they have not been widely used or proven to enhance the value of intrapartum fetal surveillance. This article discusses future directions for novel intrapartum fetal surveillance systems that leverage the long experience gained from EFM to enhance the level of risk assessment and prognosis.
Highlights
- •
Standard electronic fetal monitoring (EFM) is a limited prognostic tool.
- •
Adjunctive assessment tools have not consistently improved EFM performance.
- •
Future surveillance systems are possible with today’s technologies.
- •
Such systems will need to blend the existing EFM systems with intelligent computers.
- •
Industry must be willing to take on the risk of developing such novel systems.
Introduction and background
Electronic fetal monitoring (EFM) is entering its fifth decade in intrapartum care. In spite of the best of intentions, standard EFM using unaided visual interpretation of fetal heart rate (FHR) patterns has yet to demonstrate improved perinatal outcomes such as lower rates of perinatal mortality or morbidity . Recognized shortcomings of EFM include, but are not limited to, reliability and reproducibility of FHR pattern recognition as demonstrated by numerous studies comparing agreement levels for FHR interpretation among experienced clinicians .
In order to begin addressing some of the known shortcomings of EFM, the National Institute of Child Health and Human Development (NICHD) convened an American expert panel to develop standardized terminology for visual interpretation of FHR. The deliberations of this panel, as well as those of a subsequent similar expert panel, resulted in an American College of Obstetricians and Gynecologists (ACOG) Practice Bulletin that defined a three-tier system for classifying FHR patterns . To date, no rigorous study has been performed to demonstrate the improved implementation of EFM or better obstetric outcomes by this classification scheme.
Recognizing that the best opportunity to improve the effectiveness of EFM might begin with taking the “human factor” out of the equation, efforts to develop automated systems for FHR pattern recognition have been ongoing since the 1970s . With improved computer operating systems, software, and firmware, the development of such systems has continued , leading to mature products that are now being vended and used in selected obstetric units around the world. Fig. 1 shows a typical block diagram architecture of such an analytic system. Concomitant with the appearance of such sophisticated analytic systems has been the development of a new generation of noninvasive transducers for obtaining both fetal and maternal signals . The potential benefit of more sensitive and accurate surface probes is a reduction in the “noise” of the system, thereby improving the quality of signal processing and analysis. Here, a question arises: Does the use of automated FHR analysis lead to better perinatal outcomes? A large randomized controlled trial, INFANT, was launched in the United Kingdom and Ireland ( http://www.ucl.ac.uk/cctu/researchareas/womenshealth/infant ) several years ago. Anticipated enrollment was to be 46,000 patients randomized to either a decision-support arm (automated FHR alerts available) or a control arm (alerts not available). To date, the final results of this trial have not been published.
It may well have been short-sighted to consider that automation of FHR interpretation alone would improve intrapartum care and provide healthier infants. In order to overcome the historic limitations of unaided use of EFM, numerous attempts have been made to implement adjunct technologies, including fetal scalp blood sampling, fetal pulse oximetry, and, most recently, fetal electrocardiogram (ECG) ST-segment analysis . The latter technology has shown much promise in European studies , but a large US randomized controlled trial failed to show similar positive results .
Future perspectives in fetal surveillance systems
In order to forecast the future of intrapartum fetal surveillance, one must build on the large collective experience of decades of clinical investigation. Fig. 2 shows the developmental stages of fetal surveillance systems as their degree of sophistication is increased. In Stage 1, the fetal monitor presents the raw data of the FHR and uterine activity to clinicians who must proceed to interpret these data and apply their interpretations to subsequent intrapartum care. In Stage 2, the fetal monitor provides automated FHR interpretation and/or higher-level alerts ( Fig. 1 ) that will be incorporated in fetal status assessment and prognosis. In Stage 3, the fetal monitor combines additional clinical information about the patient with the current status of the automated alerting system and proposes a potential care plan for the patient. Stage 1 was accomplished at the introduction of EFM. Stage 2 has been more recently accomplished and such monitoring systems exist and have been used for intrapartum surveillance in selected obstetric units. Stage 3 is the next horizon for future-generation electronic fetal monitors, and possible features of such systems will occupy the remainder of this monograph.
What might we anticipate in future intrapartum surveillance systems? As mentioned earlier, electronic fetal monitors were originally developed to act as decision-support tools. However, in so doing, undue emphasis was placed on FHR pattern interpretation alone. Advanced decision-support systems would need to leverage what is now known about the prognostic value of specific FHR patterns that would be imposed on large knowledge bases, as will be described later. The systems of the future will need to be user-friendly and display the traditional EFM elements, but add prognostic information that goes beyond basic alerting.
Silent alert systems
Obstetrical units are often noisy and chaotic places where the conditions of individual patients are subject to sudden and unpredicted change. Current monitoring systems link numerous individual devices to central stations that provide a continuous array of FHR data. This can not only be an overwhelming setting, but also the one that bombards care providers with large quantities of information that are often not clinically important. One possible solution is taken from a model that has been used in the banking business for decades: the silent alarm. Fig. 3 analogizes how such systems might function in a busy labor unit. The use of this model would keep fetal monitoring systems operating in the background until a situation arose where critical alerts would need to be triggered. The existing telemetry devices, enabling the physical monitors to reside outside of patient rooms, could support these “silent” systems. A potentially important downstream benefit from silent systems would be the measureable reduction in stress levels for both patients and their caregivers who would remain undistracted by the often needless intrusion of technologies.
Intelligent intrapartum surveillance systems
Given the constantly changing intrauterine environment that characterizes active labor, the ability to evaluate fetal condition and predict its course is currently well beyond the capability of conventional monitoring systems. Intrapartum conditions are not unlike those that are addressed routinely by the US National Oceanic and Atmospheric Administration (NOAA) ( http://www.noaa.gov ). Equipped with a series of environmental satellites, courtesy of the US National Aeronautics and Space Administration, NOAA has banks of intelligent computers that can predict natural disasters, and, in some instances, render alerts far enough in advance to enable mass evacuations that could save thousands of lives. Key features of these predictive and alerting systems include the ability to identify and analyze changes in regular climate patterns that are associated with severe weather events. At the backbone of such systems are hybrid software programs that employ algorithms, expert systems, and neural networks.
In an attempt to deal with all of the environmental variables that occur during labor, there have been various efforts to develop computerized solutions using any or all of the aforementioned methods . Although the mathematical formulas of algorithmic programs can address individual elements present during the course of labor, they are readily overwhelmed when presented with the massive number of varying data points that are in constant flux. Expert systems are often rule-based and, unlike algorithm-based systems, contain rules capable of recognizing a number of events that occur in real time. In fact, a number of the automated FHR interpretative programs have used expert systems to recognize FHR patterns and classify them . However, expert systems come up short when they are presented with events or conditions for which the rules are lacking or insufficient.
Neural networks are a form of “intelligent” software that, rather than using algorithms or rules, look for associations among variables via an application of matrix algebra and derive solutions that can determine the probability of events of interest. Unlike expert systems, neural network solutions have the capacity to continuously improve through “learning” by the means of increasing numbers of case examples. The drawback of neural networks is that they do not readily identify the importance of individual variables by measuring the magnitude of their effects. Although there have been relatively few examples of neural networks developed to solve obstetric problems, our group successfully piloted such an approach to prediction of perinatal outcome associated with nonstress testing ( Fig. 4 ) . Of the model systems that are described below, the computerized solutions will most likely incorporate hybrid programs that may utilize more than one of the approaches described above.
Rapid response systems
As in automobile travel, serious accidents can also happen in an obstetrical unit. Fortunately, such accidents are infrequent, but when they do occur, they can prove life threatening to both mother and child. Although not every intrapartum accident can be predicted far in advance, there are circumstances in which the events that precede them could be detected, for example, progressively worsening umbilical cord constriction that might ultimately lead to respiratory acidosis and even intrauterine death, if not relieved. Monitoring systems capable of modeling such events should provide sufficient warning in the same manner as conditions that precede automobile collisions, so that preemptive action, such as patient repositioning or amnioinfusion, could be undertaken ( Fig. 5 ).
