Computerized analysis of cardiotocograms in clinical practice and the SisPorto^® system thirty-two years after: technological, physiopathological and clinical studies

An initial will for improvement

The idea of the use of computers in CTG analysis was almost contemporary to the initial development and dissemination of cardiotocography in clinical practice, in the decades of 1970 and 1980, as a will for improvement of the technique [9, 10]. However, CTG analysis was disseminated in clinical practice, with a wide acceptance of conventional visual analysis of tracings, without a significant development of cCTG [11]. At that time, the cost, scarcity, dimensions, processing speed and programming complexity of the computers were major obstacles to the development of cCTG [12]. In that way, visual estimation of the FHR baseline, decelerations, accelerations, short and long-term variability, without cCTG, became the standard way of assessing CTGs, both in the ante and intrapartum periods [11].

A compelling challenge

In the final years of the decade of 1980, cCTG became a compelling challenge to overcome the evidence that visual analysis of CTGs was poorly reproducible [13], [14], [15], with discrepant results of sensitivity and specificity, in the detection and prevention of the same or different outcomes [16, 17]. Not surprisingly, the results of the meta-analysis studies, combining results from randomized controlled trials (RCT) that had used methods with discrepant sensitivities and sensitivities, did not evidence clear differences between visual analysis of CTGs and other methods, such as intermittent auscultation (Table 1) [5, 18], [19], [20]. In this setting, a meta-analysis by Grivel et al. [5] showed that in the antepartum period visual CTG analysis vs. no CTG analysis showed no significant difference in perinatal mortality (2.3% vs. 1.1%, 4 studies, n=1,627) or preventable deaths, though the meta-analysis was underpowered. In the same line, there was no significant difference in caesarean sections (19.7% vs. 18.5%, 3 trials, n=1,279). There was also no significant difference for 5 min-Apgar scores <7 or admission to neonatal intensive care units [5]. Similarly, the meta-analysis published by Neilson et al. [19] and Alfirevic et al. [20], regarding studies performed in the intrapartum period, showed that visual CTG analysis during labor was associated with reduced rates of neonatal seizures, but no clear differences in cerebral palsy, infant mortality or other outcomes. Moreover, continuous CTG analysis was associated with an increase in caesarean sections and instrumental vaginal births, even when there was access to fetal blood sampling [20]. On the other hand, ST waveform analysis made no obvious difference to caesarean sections (6 trials, n=26,446), severe metabolic acidosis or neonatal encephalopathy [19]. There were, however, fewer fetal scalp samples, although the findings were heterogeneous. Moreover, there were marginally fewer operative vaginal births but no obvious difference in low 5 min-Apgar scores or admissions to special care units [19]. There was little evidence that PR interval analysis conveyed any benefit [19].

Physiological and physiopathological background–the importance of the FHR baseline

Medical decisions are based on the knowledge of physiopathology and in probabilities. Accordingly, it is reasonable to develop systems for cCTG based not only on the analysis of mathematical variables used in the calculation of probabilities but also on physiopathological models translated from experimental studies understandable by clinicians [22].

Thus, it is important that cCTG systems are developed as to allow clinicians to analyse what they are familiar with. Namely the dynamics of the sympathetic and parasympathetic system in response to hypoxic, haemodynamic or other stimuli, by the identification of the FHR baseline, accelerations, decelerations and short and long-term variability [11, 22]. That will enable clinicians to anticipate and establish diagnosis, prognosis and therapeutic actions, not only based on the analysis of probabilities, performed either by humans or computers, but also based on the knowledge of what happens in the physiopathological processes.

In this setting, the definition of a measurable FHR baseline with a physiopathological meaning is crucial. Not only because of the intrinsic physiological and physiopathological meaning of the baseline but also because it is a pre-requisite for the estimation of all the other CTG variables, namely the accelerations and decelarations [11, 22]. In this way, the definition of an objective FHR baseline, early became the main challenge of the development of the Porto system/SisPorto^® for cCTG [6, 23], [24], [25]. A FHR baseline with a physiological and physiopathological meaning was developed, rather than an abstract mathematical concept or an equivocal concept interdependent of the definition of the accelerations and decelerations [26, 27]. A FHR baseline indexed to the basal FHR obtainable during fetal rest, that is, during calm fetal sleep (fetal behavioral state 1F) in the absence of fetal movements, uterine contractions, drug actions or other abnormal stimuli [26, 27]. A FHR baseline indexed to the mean FHR value of a FHR pattern A, corresponding to the cardiac intrinsic cronotropic activity under the influence of a resting autonomous and central nervous system [27]. That FHR baseline is not only meaningful from the physiological and physiopathological point of view, but is also measurable in the most reproducible way ever reported in CTG analysis [27]. Then, from that FHR baseline, accelerations, decelerations and variability may be objectively estimated [23, 24, 28, 29]: accelerations corresponding to the activation of the central nervous system (fetal behavioral states) or to the response of the sympathetic branch of the autonomous nervous system to the stimulation of the baro or chemo receptors; decelerations corresponding to the activation of the parasympathetic branch of the autonomic nervous system via the baro or chemo receptors or to a depression or blockade of the intrinsic cardiac activity; and variability corresponding to the permanent balance action of the central and the autonomic nervous system over the cardiac activity [30], [31], [32]. Figure 1 displays a cCTG performed by the Omniview-SisPorto^®, with simultaneous recording of the maternal heart rate (MHR) and FHR. At the bottom of Figure 1, the results of the “Last hour analysis” are presented, with indication of: the signal loss and quality; the basal MHR and FHR (right columns of the results) and the MHR and FHR baselines (indexed to the basal MHR and FHR–left columns of the results); the accelerations, the decelerations and the abnormal short (STV) and long term variability (LTV); and the uterine contractions (UC).

Equipment, signals acquisition, storing and retrieving

The development and dissemination of conventional fetal monitors with facilities for the communication with user-friendly low-cost personal computers were, and are, essential for the development of cCTG [1, 6, 21]. Initially the communication between fetal monitors and personal computers was established via digital ports and cables [1, 6, 21]. Now everything is more flexible with local and remote wireless intra and internet facilities.

Modern conventional fetal monitors convey maternal and fetal signals in a digital format. MHR may be acquired via electrocardiographic (ECG) or photoplethysmography (PPG)/oximetry sensors, while maternal temperature and blood pressure may be also acquired (Figure 1). On the other hand, FHR may be acquired with external or internal ECG sensors or external ultrasound probes (US). UC may also be acquired with internal sensors or external tocodynamometry [6], [7], [8, 33], [34], [35].

In Table 2, the most representative studies performed with the SisPorto^® system in relation to technological developments are summarized.

It is important to consider the kind of sensors that are used to get the maternal and fetal signals. Nunes et al. observed a higher signal loss with external monitoring with US compared to internal ECG sensors (10% vs. 4%; p<0.001) [36]. However, no differences were found in mean FHR baseline (129 bpm vs. 130 bpm, p=0.245), though more accelerations (12 vs. 8, p<0.001) and less decelerations (8 vs. 10, p<0.001) were detected with external monitoring [36]. One the other hand, Gonçalves et al. found that MHR variability indices were significantly different when obtained with ECG or PPG/oximetry sensors, with high disagreement, for entropy and fast oscillation indices, and low disagreement, for the mean MHR and slow oscillation indices [37]. However, both the acquisition modes evidenced comparable auROC values in the prediction of neonatal acidemia and operative vaginal delivery [37] (Table 2).

It is also important to consider how maternal and fetal signals are sampled by the computer from the fetal monitor digital ports. According to Gonçalves et al., using a scalp electrode, beat-to-beat sampling provided significantly better results in the estimation of FHR linear indices and 4 Hz sampling better results in entropy indices, regarding the discrimination between fetuses born with different umbilical artery blood (UAB) pHs [38].

Preprocessing algorithms

After the maternal and fetal signals acquisition by the computer, it is important to pay attention to the preprocessing algorithms. They correct errors, remove noise and, if possible, reduce the number of signals to be processed, as to make their storing, processing and display more efficient. The preprocessing algorithms are also important to provide information about signals loss and quality [6], [7], [8, 24, 33], [34], [35]. However, as the preprocessing algorithms may significantly change the original signals, it is essential to know their possible effect on the final signals processing results.

There are several studies about these less published issues, like a paper by Gonçalves et al. who verified that the resampling at 2 Hz of FHR signals acquired at 4 Hz can significantly affect most FHR variability indices [39].

Another important preprocessing issue pertains to the identification and correction of MHR sequences misrecorded and misinterpreted as FHR sequences, as to avoid severe clinical errors, including fetal deaths [40]. In this setting, Pinto et al. verified that during the final hours of labor of 61 laboring women, 72% of the FHR tracings exhibited episodes of major MHR-FHR ambiguities. The removal of MHR-FHR ambiguities resulted in a significant decrease in the detection of FHR decelerations and an improvement in FHR tracing classification, regarding the prediction of the newborn UAB pH [41] (Table 2).

Processing algorithms following the FIGO guidelines for fetal monitoring

The use of processing algorithms for CTG analysis closely following international guidelines is important to make the systems for cCTG understandable by the health professionals that are familiar with conventional CTG analysis. This has been the way followed since the beginning of the development of the SisPorto^® system that closely follows the FIGO guidelines for fetal monitoring [6], [7], [8, 11, 22, 33], [34], [35]. In this setting, the most difficult part of the development of the SisPorto^® algorithms pertained to the algorithm for the FHR baseline estimation. The FHR baseline is easy to estimate in stable FHR patterns but very difficult to ascertain in unstable FHR patterns, like largely decelerative or accelerative patterns, namely during the fetal behavioral state F4 with FHR pattern D or during second stage of labor [23, 24]. In this way, a complex algorithm, that is described in detail elsewhere [6], [7], [8, 23, 24] was developed. Then, algorithms for the detection and classification of FHR accelerations, deceleration and uterine contractions, were developed, as well as for the estimation of FHR short-term (STV) and long-term variability (LTV) [6], [7], [8, 23, 24, 33], [34], [35] (Table 2 and Figure 1).

Processing algorithms with non-conventional analysis

Modern personal computers allow the calculation of an almost endless number of mathematical non-linear and linear time and frequency domain CTG indices, and their combined use in the prediction of clinical outcomes using multivariate analysis and artificial intelligence tools [42, 43]. Some of those mathematical and computational tools have also been used with SisPorto^®, namely linear time and frequency domain indices, and non-linear entropy and fractal indices, as well as combined MHR-FHR analysis and FHR analysis using neural networks [25, 44], [45], [46], [47], [48] (Table 2).

Interface with the users, signals display and printing

The interfaces of the systems for cCTG have been assumed by their developers as essential for their acceptance and efficacy [1, 6], [7], [8]. The SisPorto^® system automatically starts signals acquisition and display, as soon as any fetal monitor connected to the system is turned on (Figure 1). The latest Omniview-SisPorto^® version provide the online display of up to 16 CTGs in a central monitoring system station emulating conventional colored CTG paper, with its characteristic scales and speeds. Alarms may be set-up at the best convenience and tracings may be printed in the conventional cardiotocographic format [8, 33].

Applications in physiological and pathophysiological studies – fetal behavioral states, fetal development and gender, twin-to-twin and maternal-fetal interactions

Besides the identification of the basic CTG features (Figure 1), the systems for cCTG should be able to identify and characterize other physiological or physiopathological variables and processes.

In Table 3, a summary of the most representative studies performed with the SisPorto^® system are presented, in relation to the fetal behavioral states, fetal development and gender, as well as twin-to-twin and maternal-fetal interactions [45], [46], [47], [48], [49], [50], [51], [52], [53], [54].

Gonçalves et al. showed that the FHR patterns associated to active sleep (pattern B) and wakefulness (pattern D) evidenced more signs of autonomous nervous system activity, with sympatho-vagal imbalance, and less signs of complexity than the patterns associated with calm sleep (pattern A) and wakefulness (pattern C) [45]. They also evidence that in fetuses with normal outcomes, from weeks 25–40 of gestational age, the mean FHR decreased significantly, whereas most variability indices increased; sympatho-vagal imbalance exhibited two local maxima at 29–30 and 34–35 weeks and decreased afterwards [46]. Additionally, Bernardes et al. evidenced that normal term female fetuses exhibited significantly more linear and significantly less complex FHR activity than their male counterparts, in the antepartum period [47]. They also showed that term female fetuses exhibited significantly more linear FHR activity than their female counterparts, while maintaining similar complex activity, in the final minutes of labor and when were born with acidosis [48] (Table 2).

Amorim-Costa et al. published reference charts for CTG parameters following the FIGO guidelines throughout normal pregnancy [49, 50]. From 24 to 41 weeks, the charts evidenced a decrease of FHR baseline and an increase of accelerations and variability; decelerations were practically zero for all gestational ages. FHR baselines were consistently higher in females whereas variability was lower [49, 50]. They also reported that during pregnancy the FHR baseline showed a more pronounced decrease in SGA fetuses, being higher at earlier gestational ages and lower later. Average LTV was significantly lower in SGA<p3 fetuses, but a parallel increase occurred in all groups. There was considerable inter-fetal variability within each group [51] (Table 3).

Studies on the influence of fetal presentation on FHR variability, maternal-fetal attachment [52] and twin-to-twin interactions [53, 54] were also published (Table 3).

Clinical studies

Table 1 summarizes the main systematic reviews about the present state of CTG analysis and cCTG in clinical practice [4, 5, 19, 20, 55, 56] and Table 4 summarizes the most representative clinical studies performed with the SisPorto^® system for cCTG [52, 57], [58], [59], [60], [61], [62], [63], [64], [65], [66], [67].

Most clinical studies on cCTG, namely those performed with the SisPorto^® system, are small validity case series studies, pertaining to the detection of compromised fetuses in high-risk pregnancies, during the antepartum period [57], [58], [59], [60], [61], or the to the detection of fetal hypoxia or low Apgar scores, during the intrapartum period [52, 62], [63], [64], [65]. Some clinical studies are cohort studies or RCT, but only a minority of them has included a large number of cases [66, 67] or was included in meta-analysis studies [4, 5, 19, 20, 55, 56].

The main clinical study performed in the antepartum period with the SisPorto^® system was published by Ayres-de-Campos et al. [59] Three hundred forty five CTGs were acquired in the same number of fetuses, from eight tertiary care centers in Europe and Australia, 4 h before an elective cesarean section. The receiver-operator curve (auROC) regarding the prediction of the 1 and 5 min Apgar scores under 5 and 7 were, respectively, 0.96–1.00 and 0.81–0.89, considering the different CTG variables, whereas the best auROC regarding the prediction of an UAB pH<7.15 was 0.69 [59] (Table 4).

The main clinical studies performed in the intrapartum period with the SisPorto^® system were published by Nunes et al. [66] and Lopes-Pereira et al. [68]. Nunes et al. published a RCT that has involved 7,735 cases from five hospitals in the United Kingdom, pertaining to singleton, vertex fetuses of 36 weeks of gestation or greater during labor [66]. The cases were randomized to continuous central fetal monitoring by cCTG and online alerts (experimental arm) or visual analysis of CTGs (control arm). There were 16 cases of metabolic acidosis (0.40%) in the experimental arm and 22 (0.58%) in the control arm (relative risk 0.69 [0.36–1.31]) with no statistically significant differences found in the incidence of secondary outcomes. Computer analysis with real-time alerts did not significantly reduce the rate of metabolic acidosis or obstetric intervention. A lower-than-expected rate of newborn metabolic acidosis was observed in both arms of the trial [66]. On the other hand, Lopes-Pereira et al. compared two cohorts, one with 8,791 cases and the other with 29,675 cases, obtained, respectively, before and after the introduction in the clinical practice of a tertiary care hospital of the Ominiview-SisPorto^® system with cCTG and ST analysis. The system was associated with significant reductions in hypoxic-ischemic encephalopathy and cesarean deliveries [68].