Research progress in intelligent control of intrapartum oxytocin (unrevised version)

Abstract With the development of precision medicine and artificial intelligence, the infusion of many drugs has been intelligently controlled according to patients’ conditions. However, the infusion of oxytocin (OT) still relies on medical staff responsible for implementing artificial regulation based on observation of fetal electronic monitoring and other maternal and fetal conditions. In this review, we discussed recent trends in intelligent infusion systems, the development status and dilemma of intelligent control of OT infusion, the drug intelligent feedback control system principle, and current risks and challenges to further promote obstetric informatization.


Current status of intrapartum OT medication
Oxytocin (OT) is a hormone produced in the hypothalamus and released into the bloodstream by the pituitary gland.Previous studies have suggested that OT can regulate myometrial smooth muscle excitability by OT binding to specific OT receptors of uterine smooth muscle [1], which can increase the uterine smooth muscle tension and the contraction frequency, maintaining rhythm, symmetry, and polarity, thereby facilitating delivery [2,3].Studies have suggested that whether OT is used to induce labor in patients with cervical ripening during pregnancy or to induce labor in parturients with prolonged labor based on painless delivery [4][5][6], it can accelerate labor, promote natural childbirth, and ensure the safety of newborns.In addition, the oxytocin challenge test (OCT) is often performed in prenatal fetal heart monitoring in clinical practice to improve the accuracy of fetal distress diagnosis [7].However, the individual sensitivity of OT clinical application tends to vary greatly.If the used dosage is too large, it may cause adverse reactions, including uterine over-efficiency, high frequency of contractions, and fetal heart decelerations [8]; in severe cases, it may lead to tetanic contractions, uterine rupture, fetal distress, and even death, with disastrous consequences for mothers and infants [9,10].Therefore, real-time, accurate, and effective OT regulation is urgently required in clinical practice.
OT is commonly administered by intravenous infusion, with an infusion pump precisely regulating the drug dose per minute in the maternal body.The Chinese Medical Association of Obstetrics Group (CMAOG) advises continuous electronic fetal monitoring and manual supervision by midwives for parturients using OT during labor to regulate the rate of OT intravenous infusion in real-time according to maternal contraction reaction and fetal heart rate, etc [11], thereby ensuring the safety and effectiveness of OT infusion.This manual judgment of the OT infusion rate is susceptible to subjective factors of midwives, such as clinical experience, personality, etc [12,13].and high labor costs [14].The latest data suggested that the number of deliveries in China reached 14.65 million in 2019 [15], and the proportion of hasten parturition and labor induction using OT reached 45.2% [16].WHO suggests that each midwife, which is uniquely able to provide essential services to women and newborns, provides services for 175 pregnant and lying-in women annually [17], while the ratio of midwives to parturients in midwifery medical institutions in China is only 1:4000, indicating a major shortage of midwifery human resources in this country [18].Considering that one midwife often monitors multiple parturients simultaneously, which leads to the failure of timely adjustment of OT intravenous infusion rate, further threatening maternal and child safety.
According to the above, clinical work currently faces the following problems: (1) the maternal population base is large, and the demand for OT infusion is growing; (2) the OT medication regulation process is cumbersome, so given the current surge in delivery volume and shortage of midwives, the infusion rate often cannot be timely adjusted, which will increase the risk of mother and child; (3) due to the influence of midwives' clinical experience, personalities, and medication habits, OT drip rate adjustment is prone to subjective judgment errors.Therefore, many parturients require more intelligent protocols to ensure the safety and accuracy of OT intravenous infusion during maternal delivery.

The present situation of the drug intelligent feedback control system
With the growth of big data and internet technology, the wisdom medical service system has become a new trend in clinical work [19], and the advantages of artificial intelligence as its important branch in nursing work.are ever more exceptional.The systems, represented by an intelligent terminal, have made great progress in medical nursing, such as drug preparation, infusion check, intelligent transfusion, and health management [18].An intelligent infusion system refers to the process that the mobile monitoring terminal monitors the patient's basic vital sign parameters and feeds back the signal to the central processor in time.The central processor analyzes, automatically judges, makes decisions, provides further feedback, controls the infusion pump, and adjusts the infusion rate or the concentration of the liquid.The system can accurately control the infusion rate and realize the purpose of intelligent drug administration according to the patient's individual condition.At present, it is widely used in the intelligent control of drugs such as anesthesia and insulin [20,21].It includes the sensor, controller (central processor), and actuator (micro infusion pump), and through the connection of these three major components, an open-loop control system and a closed-loop control system can be formed [22].

Open-loop control system
The open-loop control system means that the output of the actuator (infusion pump) does not affect the controller's output (central processor).The schematic diagram is shown in Figure 1.In this control system, no closed loop is formed [23].The common clinical intelligent infusion systems, such as intelligent infusion monitors, infusion micropump, etc [24], can accurately control the infusion rate and monitor the allowance.When problems, such as liquid blockage or lack of infusion liquid, are detected, the system generates an automatic alarm [25].Because the research technology is relatively mature and connected with the hospital information system to set up an integrated infusion network, it is useful for real-time intelligent monitoring of medication.Yet its cost is still high.So far, several large hospitals have already applied this system, mainly for monitoring of critically ill patients [26].

Closed-loop control system
The closed-loop control system is characterized by the fact that the output of the actuator (infusion pump) will inversely affect the input of the controller (central processor), forming one or more closed loops [27].The schematic diagram is shown in Figure 2. Two types of closed-loop control systems are currently available: positive feedback and negative feedback.If the feedback signal is opposite to the given signal of the system, it is called negative feedback; if the polarity is the same, it is called positive feedback.General closed-loop control systems use negative feedback, also known as negative feedback control systems [28,29].De Boc et al. [30] developed a portable closed-loop insulin pump.By setting the expected plasma insulin concentration and obtaining the current subcutaneous and tissue interstitial insulin concentration in real-time, the injection value and time of each insulin injection are calculated and controlled by the control unit installed in the injection pump, which keeps blood glucose values in the body stable.
The anesthetic infusion system designed with the closed-loop feedback principle is also more accurate.The anesthesiologist sets the target efficacy as the feedback variable, and the control algorithm is based on the difference between the measured patient efficacy and the target efficacy.The drug delivery system calculates the new infusion rate or drug concentration, and individual administration is in line with clinical needs [31].Hemmerling et al. [32] developed the world's first fully automated anesthesia infusion system, McSleepy.Briefly, the patient's basic information is entered before using the system, including age, sex, height, weight, and various indicators that can be collected during the operation.From induction to maintenance, the system can automatically adjust the depth of sedation, the level of analgesia and muscle relaxation to control the process of anesthesia, so that anesthesiologists can release more time to monitor the condition of patients.

The principle of intelligent control of OT during delivery
The existing research has applied computer technology to the intelligent regulation of OT.The main principle involves coneecting electronic fetal monitors (EFMs), infusion pumps (IPs), and a central processor through wireless transmission.The changes in uterine contraction pressure and fetal heart rate are measured through the external uterine contraction probe and ultrasonic Doppler probe of a fetal heart monitor.The fetal heart contraction signal is transmitted to the central processor for summary and then analyzed by the internet of things technology.Finally, the central processor sends out regulation instructions to control the bedside infusion micropump of the parturient to control the injection speed.At the same time, the infusion pump transmits the real-time infusion speed and dose of OT to the central processor as a basis for regulation and decision-making.The principle of a closed-loop control system is adopted to realize the real-time adjustment of OT infusion speed according to fetal heart rate and uterine contraction (Figure 3).

Development status of intelligent regulation and control of OT during delivery
In the aspect of software development, HuTingting et al. [33] proposed an intelligent OT dose control method to extract data such as uterine cavity pressure, uterine contraction frequency, duration of uterine contraction, fetal heart rate alone or combined with demographic data and nursing records from uterine contraction signals.The dose prediction model of OT during delivery was constructed based on Bayesian optimized LightGBM algorithm.A total of 10061 OT regulation records were included in this study.The prediction accuracy, precision, recall rate, and F1 value of the model were 0.83, 0.853, 0.828, and 0.84, respectively.The results show that the prediction system based on machine learning can simulate human brain thinking, read the contraction signal and automatically adjust the dose of oxytocin by learning the situation of a large number of midwives regulating the dose of oxytocin.The system can be embedded in the central processor as a data analysis center to achieve intelligent regulation of infusion micropump, forming a regulation system that is convenient for monitoring uterine contraction and fetal heart rate changes, intelligent regulation of OT infusion, and manual monitoring.
In recent years, the number of vaginal births after cesarean section (VBAC) has increased [34], and the demand for VBAC people to use OT to induce labor is gradually increasing.However, because the previous cesarean section destroyed the continuity of the myometrium in the lower segment of the uterus, VBAC parturients should be more cautious than ordinary parturients in the infusion of OT.Therefore, some scholars have proposed an intelligent OT regulation method for this population [35], which includes realtime analysis of fetal heart monitor information (such as duration of uterine contraction, intrauterine pressure, fetal heart rate, frequency of uterine contraction), maternal body mass index (BMI), interval time between previous cesarean section, history of spontaneous delivery, labor analgesia and so on.The intelligent regulation model of oxytocin in the vaginal trial delivery population after the cesarean section was constructed based on the XGBoost algorithm.The accuracy of the model was 0.82, and the recall rate was 0.82.The clinical verification method involves selecting 10 samples and predicting their speed regulation for 10 times.In manual regulation, a low seniority midwife (i.e. less than 5 years of working experience with a junior professional title) independently judged fetal heart and contractions and implemented OT drip rate regulation.The decision was made by two senior experts to manually correct the OT control records, and this is the gold standard.If the experts have different opinions on the decision, they will give their opinions after joint discussion, and compare the differences between the decision results of the artificial adjustment mode and the prediction model and the experts' opinions, and express the accuracy rate (%).The result showed that the accuracy of the prediction model is close to the level of expert decision making and better than the regulation and control decision of low seniority midwives.
Mariana et al. simulated the interaction relationship between uterine contraction frequency, uterine cavity pressure, cervical dilatation, fetal presentation, and OT concentration in the human body based on OT pharmacokinetic methods and automatically adjusted the OT dose according to real-time changes in input variables [36].Based on the principle of the metabolic rate of OT in the human body, this method stimulates the process of human medication and metabolism, further verifies the feasibility of accurately regulating the infusion rate of OT through the individual response of pregnant women, and provides theoretical support for the next clinical practice.
Regarding hardware development, Xiaoyu et al. designed an oxytocin automatic drip regulation device for obstetrics [37], the first reported study on OT intelligent control.In his study, the normal range data of contractions, device parameters, and other relevant data are set through the human-computer interface, and the frequency and intensity signals of contractions are collected through the pressure sensor.The CPU unit of the DSP chip receives the signals and compares them with the preset normal range data of contractions.If data are different from the normal contraction frequency and intensity, the control signals are output through the signal amplification and driving circuit to the infusion pump, which then regulates the dripping speed of the liquid medicine to the normal range.This study elaborated on the collection and transmission process of uterine contraction signals and put forward the concept of precise adjustment of OT infusion rate through individual maternal response, which laid a theoretical foundation for further study.
Aiyu et al. [38] designed a set of the infusion system software, including a monitoring device, a computer system, and an OT dripping speed control device.The maternal state is monitored by a uterine contraction pressure sensor and a fetal heart Doppler sensor and fed back to the computer system, which controls the dripping speed of the oxytocin based on the maternal state.In addition, the system can send the status of contractions and oxytocin instillation to medical staff or maternal relatives in time.At the same time, it also has the capacity for self-learning.When the accidental misoperation of medical staff and the input set value do not match the uterine contraction frequency, uterine cavity pressure, and fetal heart rate, appropriate reminders and confirmation will be made to avoid accidental errors of medical staff.
In summary, all the cited studies have a large limitation in that electronic fetal monitoring is hugely misinterpreted by humans, and there are no validated reliable AI-driven methods for detecting patterns that accurately predict compromised versus health fetal status.This limitation is an important aspect of this work when it comes to precision automated dosing of OT.What's more, the hardware and software development of oxytocin intelligent regulation is still in its infancy.In the judgment of drug dose decisionmaking basis, most studies are simple and direct by setting the upper and lower threshold of fetal heart rate and uterine contraction, without actual signal processing; secondly, the above studies have not achieved product research, development, and transformation.Therefore, in future research, real clinical medication data should be further collected to establish a predictive model to accelerate product transformation and clinical application.

Discussion on the safety and effectiveness of oxytocin intelligent infusion control system
With the increasingly widespread use of intelligent infusion systems in clinical practice, their safety, and effectiveness have become the focus of attention [39].First, the fetal heart monitor collects uterine contractions from pregnant woman and uploads them to the central processing system through signal transmission for analysis.During this process, network-related issues such as signal interruption or delay may occur [40].Therefore, it is of significant importance to ensure the stability of fetal monitor data transmission for the safety of the oxytocin intelligent infusion system.The development of 5 G network technology provides a new opportunity for the advancement of mobile health.5 G possesses technical features such as ultrahigh data rate, massive connectivity, low latency, high reliability, etc, which has broad application prospects in the manufacturing of intelligent infusion devices [41].In future studies, the deployment of 5 G infrastructure should be expanded, and interdisciplinary collaboration should be strengthened in the development of intelligent infusion devices, to provide technical support for the intelligent regulation of oxytocin delivery.
The central processing system makes the decision of infusion rate based on the maternal uterine contractions and fetal heart rate, and the accuracy of this model is the key technology of the whole system.Currently, there is a certain gap in the accuracy of prediction of oxytocin intelligence control model, and most of them are still in the experimental stage and have not been applied in clinical practice [42].If the parameter anomaly is not accurately identified or the judgment is wrong, the condition of the puerpera may be further exacerbated, increasing her economic burden.Yet, the accuracy of the model can be improved by using appropriate algorithms to construct a prediction model, strengthening data preprocessing, and optimizing the parameter adjustment.Moreover, to improve the specificity and sensitivity of the model, the situation of reducing the rate or suspending the infusion should be regarded as the priority, followed by increasing the rate or maintaining the original rate to further ensure the safety of clinical medication.
Second, it is worth considering whether the system's decision on infusion rate should be confirmed by the medical staff and then sent to the infusion pump.Some experts believe that relying entirely on the computer for infusion regulation poses a great risk to patient medication safety, and if the system misjudges the patient information, continuing medication will further exacerbate the condition.Additionally, the complete automation of the infusion regulation process reduces the contact between nurses and patients, so some emergencies cannot be timely detected.Other experts argue that if the system sends infusion instructions after the confirmation of the medical staff, it does not completely liberate the nurses during the infusion process.Thus, a system with a reliable algorithm and high accuracy is essential for solving the problem.

Conclusion
There is still a lack of effective intelligent control in OT infusion.In light of an understanding of the influencing factors and development status quo, further study should develop the prediction model of OT infusion rate based on the artificial intelligence algorithm and large-scale sample training, provide auxiliary support for clinical decision-making of intrapartum OT administration, reduce the workload of medical staff, and achieve accurate administration.

Disclosure statement
No potential conflict of interest was reported by the authors.

Figure 1 .
Figure 1.Schematic diagram of the open-loop control system.

Figure 2 .
Figure 2. Principle of the closed-loop control system.

Figure 3 .
Figure 3. Principle of the OT intelligent control system during delivery.